diff --git a/src-tauri/src/codex_config.rs b/src-tauri/src/codex_config.rs
index 642e4886..b84caab7 100644
--- a/src-tauri/src/codex_config.rs
+++ b/src-tauri/src/codex_config.rs
@@ -9,6 +9,7 @@ use crate::error::AppError;
 use serde_json::Value;
 use std::fs;
 use std::path::Path;
+use toml_edit::DocumentMut;
 
 /// 获取 Codex 配置目录路径
 pub fn get_codex_config_dir() -> PathBuf {
@@ -135,3 +136,335 @@ pub fn read_and_validate_codex_config_text() -> Result<String, AppError> {
     validate_config_toml(&s)?;
     Ok(s)
 }
+
+/// Update a field in Codex config.toml using toml_edit (syntax-preserving).
+///
+/// Supported fields:
+/// - `"base_url"`: writes to `[model_providers.<current>].base_url` if `model_provider` exists,
+///    otherwise falls back to top-level `base_url`.
+/// - `"model"`: writes to top-level `model` field.
+///
+/// Empty value removes the field.
+pub fn update_codex_toml_field(toml_str: &str, field: &str, value: &str) -> Result<String, String> {
+    let mut doc = toml_str
+        .parse::<DocumentMut>()
+        .map_err(|e| format!("TOML parse error: {e}"))?;
+
+    let trimmed = value.trim();
+
+    match field {
+        "base_url" => {
+            let model_provider = doc
+                .get("model_provider")
+                .and_then(|item| item.as_str())
+                .map(str::to_string);
+
+            if let Some(provider_key) = model_provider {
+                // Ensure [model_providers] table exists
+                if doc.get("model_providers").is_none() {
+                    doc["model_providers"] = toml_edit::table();
+                }
+
+                if let Some(model_providers) = doc["model_providers"].as_table_mut() {
+                    // Ensure [model_providers.<provider_key>] table exists
+                    if !model_providers.contains_key(&provider_key) {
+                        model_providers[&provider_key] = toml_edit::table();
+                    }
+
+                    if let Some(provider_table) = model_providers[&provider_key].as_table_mut() {
+                        if trimmed.is_empty() {
+                            provider_table.remove("base_url");
+                        } else {
+                            provider_table["base_url"] = toml_edit::value(trimmed);
+                        }
+                        return Ok(doc.to_string());
+                    }
+                }
+            }
+
+            // Fallback: no model_provider or structure mismatch → top-level base_url
+            if trimmed.is_empty() {
+                doc.as_table_mut().remove("base_url");
+            } else {
+                doc["base_url"] = toml_edit::value(trimmed);
+            }
+        }
+        "model" => {
+            if trimmed.is_empty() {
+                doc.as_table_mut().remove("model");
+            } else {
+                doc["model"] = toml_edit::value(trimmed);
+            }
+        }
+        _ => return Err(format!("unsupported field: {field}")),
+    }
+
+    Ok(doc.to_string())
+}
+
+/// Remove `base_url` from the active model_provider section only if it matches `predicate`.
+/// Also removes top-level `base_url` if it matches.
+/// Used by proxy cleanup to strip local proxy URLs without touching user-configured URLs.
+pub fn remove_codex_toml_base_url_if(toml_str: &str, predicate: impl Fn(&str) -> bool) -> String {
+    let mut doc = match toml_str.parse::<DocumentMut>() {
+        Ok(doc) => doc,
+        Err(_) => return toml_str.to_string(),
+    };
+
+    let model_provider = doc
+        .get("model_provider")
+        .and_then(|item| item.as_str())
+        .map(str::to_string);
+
+    if let Some(provider_key) = model_provider {
+        if let Some(model_providers) = doc
+            .get_mut("model_providers")
+            .and_then(|v| v.as_table_mut())
+        {
+            if let Some(provider_table) = model_providers
+                .get_mut(provider_key.as_str())
+                .and_then(|v| v.as_table_mut())
+            {
+                let should_remove = provider_table
+                    .get("base_url")
+                    .and_then(|item| item.as_str())
+                    .map(&predicate)
+                    .unwrap_or(false);
+                if should_remove {
+                    provider_table.remove("base_url");
+                }
+            }
+        }
+    }
+
+    // Fallback: also clean up top-level base_url if it matches
+    let should_remove_root = doc
+        .get("base_url")
+        .and_then(|item| item.as_str())
+        .map(&predicate)
+        .unwrap_or(false);
+    if should_remove_root {
+        doc.as_table_mut().remove("base_url");
+    }
+
+    doc.to_string()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn base_url_writes_into_correct_model_provider_section() {
+        let input = r#"model_provider = "any"
+model = "gpt-5.1-codex"
+
+[model_providers.any]
+name = "any"
+wire_api = "responses"
+"#;
+
+        let result = update_codex_toml_field(input, "base_url", "https://example.com/v1").unwrap();
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+
+        let base_url = parsed
+            .get("model_providers")
+            .and_then(|v| v.get("any"))
+            .and_then(|v| v.get("base_url"))
+            .and_then(|v| v.as_str())
+            .expect("base_url should be in model_providers.any");
+        assert_eq!(base_url, "https://example.com/v1");
+
+        // Should NOT have top-level base_url
+        assert!(parsed.get("base_url").is_none());
+
+        // wire_api preserved
+        let wire_api = parsed
+            .get("model_providers")
+            .and_then(|v| v.get("any"))
+            .and_then(|v| v.get("wire_api"))
+            .and_then(|v| v.as_str());
+        assert_eq!(wire_api, Some("responses"));
+    }
+
+    #[test]
+    fn base_url_creates_section_when_missing() {
+        let input = r#"model_provider = "custom"
+model = "gpt-4"
+"#;
+
+        let result = update_codex_toml_field(input, "base_url", "https://custom.api/v1").unwrap();
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+
+        let base_url = parsed
+            .get("model_providers")
+            .and_then(|v| v.get("custom"))
+            .and_then(|v| v.get("base_url"))
+            .and_then(|v| v.as_str())
+            .expect("should create section and set base_url");
+        assert_eq!(base_url, "https://custom.api/v1");
+    }
+
+    #[test]
+    fn base_url_falls_back_to_top_level_without_model_provider() {
+        let input = r#"model = "gpt-4"
+"#;
+
+        let result = update_codex_toml_field(input, "base_url", "https://fallback.api/v1").unwrap();
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+
+        let base_url = parsed
+            .get("base_url")
+            .and_then(|v| v.as_str())
+            .expect("should set top-level base_url");
+        assert_eq!(base_url, "https://fallback.api/v1");
+    }
+
+    #[test]
+    fn clearing_base_url_removes_only_from_correct_section() {
+        let input = r#"model_provider = "any"
+
+[model_providers.any]
+name = "any"
+base_url = "https://old.api/v1"
+wire_api = "responses"
+
+[mcp_servers.context7]
+command = "npx"
+"#;
+
+        let result = update_codex_toml_field(input, "base_url", "").unwrap();
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+
+        // base_url removed from model_providers.any
+        let any_section = parsed
+            .get("model_providers")
+            .and_then(|v| v.get("any"))
+            .expect("model_providers.any should exist");
+        assert!(any_section.get("base_url").is_none());
+
+        // wire_api preserved
+        assert_eq!(
+            any_section.get("wire_api").and_then(|v| v.as_str()),
+            Some("responses")
+        );
+
+        // mcp_servers untouched
+        assert!(parsed.get("mcp_servers").is_some());
+    }
+
+    #[test]
+    fn model_field_operates_on_top_level() {
+        let input = r#"model_provider = "any"
+model = "gpt-4"
+
+[model_providers.any]
+name = "any"
+"#;
+
+        let result = update_codex_toml_field(input, "model", "gpt-5").unwrap();
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+        assert_eq!(parsed.get("model").and_then(|v| v.as_str()), Some("gpt-5"));
+
+        // Clear model
+        let result2 = update_codex_toml_field(&result, "model", "").unwrap();
+        let parsed2: toml::Value = toml::from_str(&result2).unwrap();
+        assert!(parsed2.get("model").is_none());
+    }
+
+    #[test]
+    fn preserves_comments_and_whitespace() {
+        let input = r#"# My Codex config
+model_provider = "any"
+model = "gpt-4"
+
+# Provider section
+[model_providers.any]
+name = "any"
+base_url = "https://old.api/v1"
+"#;
+
+        let result = update_codex_toml_field(input, "base_url", "https://new.api/v1").unwrap();
+
+        // Comments should be preserved
+        assert!(result.contains("# My Codex config"));
+        assert!(result.contains("# Provider section"));
+    }
+
+    #[test]
+    fn does_not_misplace_when_profiles_section_follows() {
+        let input = r#"model_provider = "any"
+
+[model_providers.any]
+name = "any"
+base_url = "https://old.api/v1"
+
+[profiles.default]
+model = "gpt-4"
+"#;
+
+        let result = update_codex_toml_field(input, "base_url", "https://new.api/v1").unwrap();
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+
+        // base_url in correct section
+        let base_url = parsed
+            .get("model_providers")
+            .and_then(|v| v.get("any"))
+            .and_then(|v| v.get("base_url"))
+            .and_then(|v| v.as_str());
+        assert_eq!(base_url, Some("https://new.api/v1"));
+
+        // profiles section untouched
+        let profile_model = parsed
+            .get("profiles")
+            .and_then(|v| v.get("default"))
+            .and_then(|v| v.get("model"))
+            .and_then(|v| v.as_str());
+        assert_eq!(profile_model, Some("gpt-4"));
+    }
+
+    #[test]
+    fn remove_base_url_if_predicate() {
+        let input = r#"model_provider = "any"
+
+[model_providers.any]
+name = "any"
+base_url = "http://127.0.0.1:5000/v1"
+wire_api = "responses"
+"#;
+
+        let result =
+            remove_codex_toml_base_url_if(input, |url| url.starts_with("http://127.0.0.1"));
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+
+        let any_section = parsed
+            .get("model_providers")
+            .and_then(|v| v.get("any"))
+            .unwrap();
+        assert!(any_section.get("base_url").is_none());
+        assert_eq!(
+            any_section.get("wire_api").and_then(|v| v.as_str()),
+            Some("responses")
+        );
+    }
+
+    #[test]
+    fn remove_base_url_if_keeps_non_matching() {
+        let input = r#"model_provider = "any"
+
+[model_providers.any]
+base_url = "https://production.api/v1"
+"#;
+
+        let result =
+            remove_codex_toml_base_url_if(input, |url| url.starts_with("http://127.0.0.1"));
+        let parsed: toml::Value = toml::from_str(&result).unwrap();
+
+        let base_url = parsed
+            .get("model_providers")
+            .and_then(|v| v.get("any"))
+            .and_then(|v| v.get("base_url"))
+            .and_then(|v| v.as_str());
+        assert_eq!(base_url, Some("https://production.api/v1"));
+    }
+}
diff --git a/src-tauri/src/services/proxy.rs b/src-tauri/src/services/proxy.rs
index ccb755fc..02ce4a58 100644
--- a/src-tauri/src/services/proxy.rs
+++ b/src-tauri/src/services/proxy.rs
@@ -19,7 +19,7 @@ use tokio::sync::RwLock;
 /// 用于接管 Live 配置时的占位符（避免客户端提示缺少 key，同时不泄露真实 Token）
 const PROXY_TOKEN_PLACEHOLDER: &str = "PROXY_MANAGED";
 
-/// 代理接管模式下需要从 Claude Live 配置中移除的“模型覆盖”字段。
+/// 代理接管模式下需要从 Claude Live 配置中移除的"模型覆盖"字段。
 ///
 /// 原因：接管模式切换供应商时不会写回 Live 配置，如果保留这些字段，
 /// Claude Code 会继续以旧模型名发起请求，导致新供应商不支持时失败。
@@ -52,7 +52,7 @@ impl ProxyService {
 
     /// 清理接管模式下 Claude Live 配置中的模型覆盖字段。
     ///
-    /// 这可以避免“接管开启后切换供应商仍使用旧模型”的问题。
+    /// 这可以避免"接管开启后切换供应商仍使用旧模型"的问题。
     /// 注意：此方法不会修改 Token/Base URL 的接管占位符，仅移除模型字段。
     pub fn cleanup_claude_model_overrides_in_live(&self) -> Result<(), String> {
         let mut config = self.read_claude_live()?;
@@ -1162,7 +1162,7 @@ impl ProxyService {
     ) -> Result<(), String> {
         let app_type_str = app_type.as_str();
 
-        // 1) 优先从 Live 备份恢复（这是“原始 Live”的唯一可靠来源）
+        // 1) 优先从 Live 备份恢复（这是"原始 Live"的唯一可靠来源）
         let backup = self
             .db
             .get_live_backup(app_type_str)
@@ -1181,7 +1181,7 @@ impl ProxyService {
             return Ok(());
         }
 
-        // 2.1) 优先从 SSOT（当前供应商）重建 Live（比“清理字段”更可用）
+        // 2.1) 优先从 SSOT（当前供应商）重建 Live（比"清理字段"更可用）
         match self.restore_live_from_ssot_for_app(app_type) {
             Ok(true) => {
                 log::info!("{app_type_str} Live 配置已从 SSOT 恢复（无备份兜底）");
@@ -1358,51 +1358,9 @@ impl ProxyService {
         Ok(())
     }
 
+    /// Remove local proxy base_url from TOML（委托给 codex_config 共享实现）
     fn remove_local_toml_base_url(toml_str: &str) -> String {
-        use toml_edit::DocumentMut;
-
-        let mut doc = match toml_str.parse::<DocumentMut>() {
-            Ok(doc) => doc,
-            Err(_) => return toml_str.to_string(),
-        };
-
-        let model_provider = doc
-            .get("model_provider")
-            .and_then(|item| item.as_str())
-            .map(str::to_string);
-
-        if let Some(provider_key) = model_provider {
-            if let Some(model_providers) = doc
-                .get_mut("model_providers")
-                .and_then(|v| v.as_table_mut())
-            {
-                if let Some(provider_table) = model_providers
-                    .get_mut(provider_key.as_str())
-                    .and_then(|v| v.as_table_mut())
-                {
-                    let should_remove = provider_table
-                        .get("base_url")
-                        .and_then(|item| item.as_str())
-                        .map(Self::is_local_proxy_url)
-                        .unwrap_or(false);
-                    if should_remove {
-                        provider_table.remove("base_url");
-                    }
-                }
-            }
-        }
-
-        // 兜底：清理顶层 base_url（仅当它看起来像本地代理地址）
-        let should_remove_root = doc
-            .get("base_url")
-            .and_then(|item| item.as_str())
-            .map(Self::is_local_proxy_url)
-            .unwrap_or(false);
-        if should_remove_root {
-            doc.as_table_mut().remove("base_url");
-        }
-
-        doc.to_string()
+        crate::codex_config::remove_codex_toml_base_url_if(toml_str, Self::is_local_proxy_url)
     }
 
     fn cleanup_gemini_takeover_placeholders_in_live(&self) -> Result<(), String> {
@@ -1459,7 +1417,7 @@ impl ProxyService {
         Ok(())
     }
 
-    /// 检测 Live 配置是否处于“被接管”的残留状态
+    /// 检测 Live 配置是否处于"被接管"的残留状态
     ///
     /// 用于兜底处理：当数据库备份缺失但 Live 文件已经写成代理占位符时，
     /// 启动流程可以据此触发恢复逻辑。
@@ -1675,49 +1633,10 @@ impl ProxyService {
 
     // ==================== Live 配置读写辅助方法 ====================
 
-    /// 更新 TOML 字符串中的 base_url
+    /// 更新 TOML 字符串中的 base_url（委托给 codex_config 共享实现）
     fn update_toml_base_url(toml_str: &str, new_url: &str) -> String {
-        use toml_edit::DocumentMut;
-
-        let mut doc = match toml_str.parse::<DocumentMut>() {
-            Ok(doc) => doc,
-            Err(_) => return toml_str.to_string(),
-        };
-
-        // Codex 的 config.toml 通常是：
-        // model_provider = "any"
-        //
-        // [model_providers.any]
-        // base_url = "https://.../v1"
-        //
-        // 所以接管时要“精准”修改当前 model_provider 对应的 model_providers.<name>.base_url，
-        // 避免写错位置导致 Codex 仍然走旧地址。
-        let model_provider = doc
-            .get("model_provider")
-            .and_then(|item| item.as_str())
-            .map(str::to_string);
-
-        if let Some(provider_key) = model_provider {
-            if doc.get("model_providers").is_none() {
-                doc["model_providers"] = toml_edit::table();
-            }
-
-            if let Some(model_providers) = doc["model_providers"].as_table_mut() {
-                if !model_providers.contains_key(&provider_key) {
-                    model_providers[&provider_key] = toml_edit::table();
-                }
-
-                if let Some(provider_table) = model_providers[&provider_key].as_table_mut() {
-                    provider_table["base_url"] = toml_edit::value(new_url);
-                    return doc.to_string();
-                }
-            }
-        }
-
-        // 兜底：如果没有 model_provider 或结构不符合预期，则退回修改顶层 base_url。
-        doc["base_url"] = toml_edit::value(new_url);
-
-        doc.to_string()
+        crate::codex_config::update_codex_toml_field(toml_str, "base_url", new_url)
+            .unwrap_or_else(|_| toml_str.to_string())
     }
 
     fn read_claude_live(&self) -> Result<Value, String> {
@@ -2228,7 +2147,7 @@ model = "gpt-5.1-codex"
         db.set_current_provider("claude", "a")
             .expect("set current provider");
 
-        // 模拟“已接管”状态：存在 Live 备份（内容不重要，会被热切换更新）
+        // 模拟"已接管"状态：存在 Live 备份（内容不重要，会被热切换更新）
         db.save_live_backup("claude", "{\"env\":{}}")
             .await
             .expect("seed live backup");
diff --git a/src/components/providers/AddProviderDialog.tsx b/src/components/providers/AddProviderDialog.tsx
index a40befdc..f8fc8e5b 100644
--- a/src/components/providers/AddProviderDialog.tsx
+++ b/src/components/providers/AddProviderDialog.tsx
@@ -17,6 +17,7 @@ import { UniversalProviderPanel } from "@/components/universal";
 import { providerPresets } from "@/config/claudeProviderPresets";
 import { codexProviderPresets } from "@/config/codexProviderPresets";
 import { geminiProviderPresets } from "@/config/geminiProviderPresets";
+import { extractCodexBaseUrl } from "@/utils/providerConfigUtils";
 import type { OpenClawSuggestedDefaults } from "@/config/openclawProviderPresets";
 import type { UniversalProviderPreset } from "@/config/universalProviderPresets";
 
@@ -179,11 +180,9 @@ export function AddProviderDialog({
         } else if (appId === "codex") {
           const config = parsedConfig.config as string | undefined;
           if (config) {
-            const baseUrlMatch = config.match(
-              /base_url\s*=\s*["']([^"']+)["']/,
-            );
-            if (baseUrlMatch?.[1]) {
-              addUrl(baseUrlMatch[1]);
+            const extractedBaseUrl = extractCodexBaseUrl(config);
+            if (extractedBaseUrl) {
+              addUrl(extractedBaseUrl);
             }
           }
         } else if (appId === "gemini") {
diff --git a/src/components/providers/ProviderCard.tsx b/src/components/providers/ProviderCard.tsx
index 96d8b460..3301fbc9 100644
--- a/src/components/providers/ProviderCard.tsx
+++ b/src/components/providers/ProviderCard.tsx
@@ -13,6 +13,7 @@ import { ProviderIcon } from "@/components/ProviderIcon";
 import UsageFooter from "@/components/UsageFooter";
 import { ProviderHealthBadge } from "@/components/providers/ProviderHealthBadge";
 import { FailoverPriorityBadge } from "@/components/providers/FailoverPriorityBadge";
+import { extractCodexBaseUrl } from "@/utils/providerConfigUtils";
 import { useProviderHealth } from "@/lib/query/failover";
 import { useUsageQuery } from "@/lib/query/queries";
 
@@ -76,9 +77,9 @@ const extractApiUrl = (provider: Provider, fallbackText: string) => {
     const baseUrl = (config as Record<string, any>)?.config;
 
     if (typeof baseUrl === "string" && baseUrl.includes("base_url")) {
-      const match = baseUrl.match(/base_url\s*=\s*['"]([^'"]+)['"]/);
-      if (match?.[1]) {
-        return match[1];
+      const extractedBaseUrl = extractCodexBaseUrl(baseUrl);
+      if (extractedBaseUrl) {
+        return extractedBaseUrl;
       }
     }
   }
diff --git a/src/components/providers/forms/OpenClawFormFields.tsx b/src/components/providers/forms/OpenClawFormFields.tsx
index 8c35e0df..314f7024 100644
--- a/src/components/providers/forms/OpenClawFormFields.tsx
+++ b/src/components/providers/forms/OpenClawFormFields.tsx
@@ -383,7 +383,9 @@ export function OpenClawFormFields({
                               className="flex items-center gap-1.5 cursor-pointer select-none"
                             >
                               <Checkbox
-                                checked={(model.input ?? ["text"]).includes(type)}
+                                checked={(model.input ?? ["text"]).includes(
+                                  type,
+                                )}
                                 onCheckedChange={(checked) => {
                                   const current = model.input ?? ["text"];
                                   const next = checked
diff --git a/src/components/providers/forms/hooks/useCodexCommonConfig.ts b/src/components/providers/forms/hooks/useCodexCommonConfig.ts
index fb65053d..228ab419 100644
--- a/src/components/providers/forms/hooks/useCodexCommonConfig.ts
+++ b/src/components/providers/forms/hooks/useCodexCommonConfig.ts
@@ -285,7 +285,13 @@ export function useCodexCommonConfig({
         isUpdatingFromCommonConfig.current = false;
       }, 0);
     },
-    [codexConfig, commonConfigSnippet, onConfigChange, parseCommonConfigSnippet, t],
+    [
+      codexConfig,
+      commonConfigSnippet,
+      onConfigChange,
+      parseCommonConfigSnippet,
+      t,
+    ],
   );
 
   // 处理通用配置片段变化
diff --git a/src/components/providers/forms/hooks/useOpenclawFormState.ts b/src/components/providers/forms/hooks/useOpenclawFormState.ts
index b7652570..7f32424c 100644
--- a/src/components/providers/forms/hooks/useOpenclawFormState.ts
+++ b/src/components/providers/forms/hooks/useOpenclawFormState.ts
@@ -171,18 +171,15 @@ export function useOpenclawFormState({
     [updateOpenclawConfig],
   );
 
-  const resetOpenclawState = useCallback(
-    (config?: OpenClawProviderConfig) => {
-      setOpenclawProviderKey("");
-      setOpenclawBaseUrl(config?.baseUrl || "");
-      setOpenclawApiKey(config?.apiKey || "");
-      setOpenclawApi(config?.api || "openai-completions");
-      setOpenclawModels(config?.models || []);
-      const ua = config?.headers ? "User-Agent" in config.headers : false;
-      setOpenclawUserAgent(ua);
-    },
-    [],
-  );
+  const resetOpenclawState = useCallback((config?: OpenClawProviderConfig) => {
+    setOpenclawProviderKey("");
+    setOpenclawBaseUrl(config?.baseUrl || "");
+    setOpenclawApiKey(config?.apiKey || "");
+    setOpenclawApi(config?.api || "openai-completions");
+    setOpenclawModels(config?.models || []);
+    const ua = config?.headers ? "User-Agent" in config.headers : false;
+    setOpenclawUserAgent(ua);
+  }, []);
 
   return {
     openclawProviderKey,
diff --git a/src/components/providers/forms/hooks/useSpeedTestEndpoints.ts b/src/components/providers/forms/hooks/useSpeedTestEndpoints.ts
index ab9f54b5..c503289c 100644
--- a/src/components/providers/forms/hooks/useSpeedTestEndpoints.ts
+++ b/src/components/providers/forms/hooks/useSpeedTestEndpoints.ts
@@ -3,6 +3,7 @@ import type { AppId } from "@/lib/api";
 import type { ProviderPreset } from "@/config/claudeProviderPresets";
 import type { CodexProviderPreset } from "@/config/codexProviderPresets";
 import type { ProviderMeta, EndpointCandidate } from "@/types";
+import { extractCodexBaseUrl } from "@/utils/providerConfigUtils";
 
 type PresetEntry = {
   id: string;
@@ -128,10 +129,9 @@ export function useSpeedTestEndpoints({
         }
       | undefined;
     const configStr = initialCodexConfig?.config ?? "";
-    // 从 TOML 中提取 base_url
-    const match = /base_url\s*=\s*["']([^"']+)["']/i.exec(configStr);
-    if (match?.[1]) {
-      add(match[1]);
+    const extractedBaseUrl = extractCodexBaseUrl(configStr);
+    if (extractedBaseUrl) {
+      add(extractedBaseUrl);
     }
 
     // 3. 预设中的 endpointCandidates
@@ -141,11 +141,9 @@ export function useSpeedTestEndpoints({
         const preset = entry.preset as CodexProviderPreset;
         // 添加预设自己的 baseUrl
         const presetConfig = preset.config || "";
-        const presetMatch = /base_url\s*=\s*["']([^"']+)["']/i.exec(
-          presetConfig,
-        );
-        if (presetMatch?.[1]) {
-          add(presetMatch[1]);
+        const presetBaseUrl = extractCodexBaseUrl(presetConfig);
+        if (presetBaseUrl) {
+          add(presetBaseUrl);
         }
         // 添加预设的候选端点
         if (preset.endpointCandidates) {
diff --git a/src/utils/providerConfigUtils.ts b/src/utils/providerConfigUtils.ts
index b35dcabe..84f5e941 100644
--- a/src/utils/providerConfigUtils.ts
+++ b/src/utils/providerConfigUtils.ts
@@ -1,7 +1,7 @@
 // 供应商配置处理工具函数
 
 import type { TemplateValueConfig } from "../config/claudeProviderPresets";
-import { normalizeQuotes, normalizeTomlText } from "@/utils/textNormalization";
+import { normalizeTomlText } from "@/utils/textNormalization";
 import { parse as parseToml, stringify as stringifyToml } from "smol-toml";
 
 const isPlainObject = (value: unknown): value is Record<string, any> => {
@@ -414,17 +414,234 @@ export const hasTomlCommonConfigSnippet = (
 
 // ========== Codex base_url utils ==========
 
+const TOML_SECTION_HEADER_PATTERN = /^\s*\[([^\]\r\n]+)\]\s*$/;
+const TOML_BASE_URL_PATTERN =
+  /^\s*base_url\s*=\s*(["'])([^"'\r\n]+)\1\s*(?:#.*)?$/;
+const TOML_MODEL_PATTERN = /^\s*model\s*=\s*(["'])([^"'\r\n]+)\1\s*(?:#.*)?$/;
+const TOML_MODEL_PROVIDER_LINE_PATTERN =
+  /^\s*model_provider\s*=\s*(["'])([^"'\r\n]+)\1\s*(?:#.*)?$/;
+const TOML_MODEL_PROVIDER_PATTERN =
+  /^\s*model_provider\s*=\s*(["'])([^"'\r\n]+)\1\s*(?:#.*)?$/m;
+
+interface TomlSectionRange {
+  bodyEndIndex: number;
+  bodyStartIndex: number;
+}
+
+interface TomlAssignmentMatch {
+  index: number;
+  sectionName?: string;
+  value: string;
+}
+
+const finalizeTomlText = (lines: string[]): string =>
+  lines
+    .join("\n")
+    .replace(/\n{3,}/g, "\n\n")
+    .replace(/^\n+/, "");
+
+const getTomlSectionRange = (
+  lines: string[],
+  sectionName: string,
+): TomlSectionRange | undefined => {
+  let headerLineIndex = -1;
+
+  for (let index = 0; index < lines.length; index += 1) {
+    const match = lines[index].match(TOML_SECTION_HEADER_PATTERN);
+    if (!match) {
+      continue;
+    }
+
+    if (headerLineIndex === -1) {
+      if (match[1] === sectionName) {
+        headerLineIndex = index;
+      }
+      continue;
+    }
+
+    return {
+      bodyStartIndex: headerLineIndex + 1,
+      bodyEndIndex: index,
+    };
+  }
+
+  if (headerLineIndex === -1) {
+    return undefined;
+  }
+
+  return {
+    bodyStartIndex: headerLineIndex + 1,
+    bodyEndIndex: lines.length,
+  };
+};
+
+const getTopLevelEndIndex = (lines: string[]): number => {
+  const firstSectionIndex = lines.findIndex((line) =>
+    TOML_SECTION_HEADER_PATTERN.test(line),
+  );
+  return firstSectionIndex === -1 ? lines.length : firstSectionIndex;
+};
+
+const getTomlSectionInsertIndex = (
+  lines: string[],
+  sectionRange: TomlSectionRange,
+): number => {
+  let insertIndex = sectionRange.bodyEndIndex;
+  while (
+    insertIndex > sectionRange.bodyStartIndex &&
+    lines[insertIndex - 1].trim() === ""
+  ) {
+    insertIndex -= 1;
+  }
+  return insertIndex;
+};
+
+const getCodexModelProviderName = (configText: string): string | undefined => {
+  const match = configText.match(TOML_MODEL_PROVIDER_PATTERN);
+  const providerName = match?.[2]?.trim();
+  return providerName || undefined;
+};
+
+const getCodexProviderSectionName = (
+  configText: string,
+): string | undefined => {
+  const providerName = getCodexModelProviderName(configText);
+  return providerName ? `model_providers.${providerName}` : undefined;
+};
+
+const findTomlAssignmentInRange = (
+  lines: string[],
+  pattern: RegExp,
+  startIndex: number,
+  endIndex: number,
+  sectionName?: string,
+): TomlAssignmentMatch | undefined => {
+  for (let index = startIndex; index < endIndex; index += 1) {
+    const match = lines[index].match(pattern);
+    if (match?.[2]) {
+      return {
+        index,
+        sectionName,
+        value: match[2],
+      };
+    }
+  }
+
+  return undefined;
+};
+
+const findTomlAssignments = (
+  lines: string[],
+  pattern: RegExp,
+): TomlAssignmentMatch[] => {
+  const assignments: TomlAssignmentMatch[] = [];
+  let currentSectionName: string | undefined;
+
+  lines.forEach((line, index) => {
+    const sectionMatch = line.match(TOML_SECTION_HEADER_PATTERN);
+    if (sectionMatch) {
+      currentSectionName = sectionMatch[1];
+      return;
+    }
+
+    const match = line.match(pattern);
+    if (!match?.[2]) {
+      return;
+    }
+
+    assignments.push({
+      index,
+      sectionName: currentSectionName,
+      value: match[2],
+    });
+  });
+
+  return assignments;
+};
+
+const isMcpServerSection = (sectionName?: string): boolean =>
+  sectionName === "mcp_servers" ||
+  sectionName?.startsWith("mcp_servers.") === true;
+
+const isOtherProviderSection = (
+  sectionName: string | undefined,
+  targetSectionName: string | undefined,
+): boolean =>
+  Boolean(
+    sectionName &&
+      sectionName !== targetSectionName &&
+      (sectionName === "model_providers" ||
+        sectionName.startsWith("model_providers.")),
+  );
+
+const getRecoverableBaseUrlAssignments = (
+  assignments: TomlAssignmentMatch[],
+  targetSectionName: string | undefined,
+): TomlAssignmentMatch[] =>
+  assignments.filter(
+    ({ sectionName }) =>
+      sectionName !== targetSectionName &&
+      !isMcpServerSection(sectionName) &&
+      !isOtherProviderSection(sectionName, targetSectionName),
+  );
+
+const getTopLevelModelProviderLineIndex = (lines: string[]): number => {
+  const topLevelEndIndex = getTopLevelEndIndex(lines);
+
+  for (let index = 0; index < topLevelEndIndex; index += 1) {
+    if (TOML_MODEL_PROVIDER_LINE_PATTERN.test(lines[index])) {
+      return index;
+    }
+  }
+
+  return -1;
+};
+
 // 从 Codex 的 TOML 配置文本中提取 base_url（支持单/双引号）
 export const extractCodexBaseUrl = (
   configText: string | undefined | null,
 ): string | undefined => {
   try {
     const raw = typeof configText === "string" ? configText : "";
-    // 归一化中文/全角引号，避免正则提取失败
-    const text = normalizeQuotes(raw);
+    const text = normalizeTomlText(raw);
     if (!text) return undefined;
-    const m = text.match(/base_url\s*=\s*(['"])([^'\"]+)\1/);
-    return m && m[2] ? m[2] : undefined;
+
+    const lines = text.split("\n");
+    const targetSectionName = getCodexProviderSectionName(text);
+
+    if (targetSectionName) {
+      const sectionRange = getTomlSectionRange(lines, targetSectionName);
+      if (sectionRange) {
+        const match = findTomlAssignmentInRange(
+          lines,
+          TOML_BASE_URL_PATTERN,
+          sectionRange.bodyStartIndex,
+          sectionRange.bodyEndIndex,
+          targetSectionName,
+        );
+        if (match?.value) {
+          return match.value;
+        }
+      }
+    }
+
+    const topLevelMatch = findTomlAssignmentInRange(
+      lines,
+      TOML_BASE_URL_PATTERN,
+      0,
+      getTopLevelEndIndex(lines),
+    );
+    if (topLevelMatch?.value) {
+      return topLevelMatch.value;
+    }
+
+    const fallbackAssignments = getRecoverableBaseUrlAssignments(
+      findTomlAssignments(lines, TOML_BASE_URL_PATTERN),
+      targetSectionName,
+    );
+    return fallbackAssignments.length === 1
+      ? fallbackAssignments[0].value
+      : undefined;
   } catch {
     return undefined;
   }
@@ -451,36 +668,107 @@ export const setCodexBaseUrl = (
   baseUrl: string,
 ): string => {
   const trimmed = baseUrl.trim();
-  // 归一化原文本中的引号（既能匹配，也能输出稳定格式）
-  const normalizedText = normalizeQuotes(configText);
+  const normalizedText = normalizeTomlText(configText);
+  const lines = normalizedText ? normalizedText.split("\n") : [];
+  const targetSectionName = getCodexProviderSectionName(normalizedText);
+  const allAssignments = findTomlAssignments(lines, TOML_BASE_URL_PATTERN);
+  const recoverableAssignments = getRecoverableBaseUrlAssignments(
+    allAssignments,
+    targetSectionName,
+  );
 
-  // 允许清空：当 baseUrl 为空时，移除 base_url 行
   if (!trimmed) {
     if (!normalizedText) return normalizedText;
-    const next = normalizedText
-      .split("\n")
-      .filter((line) => !/^\s*base_url\s*=/.test(line))
-      .join("\n")
-      // 避免移除后留下过多空行
-      .replace(/\n{3,}/g, "\n\n")
-      // 避免开头出现空行
-      .replace(/^\n+/, "");
-    return next;
+
+    if (targetSectionName) {
+      const sectionRange = getTomlSectionRange(lines, targetSectionName);
+      const targetMatch = sectionRange
+        ? findTomlAssignmentInRange(
+            lines,
+            TOML_BASE_URL_PATTERN,
+            sectionRange.bodyStartIndex,
+            sectionRange.bodyEndIndex,
+            targetSectionName,
+          )
+        : undefined;
+
+      if (targetMatch) {
+        lines.splice(targetMatch.index, 1);
+        return finalizeTomlText(lines);
+      }
+    }
+
+    if (recoverableAssignments.length === 1) {
+      lines.splice(recoverableAssignments[0].index, 1);
+      return finalizeTomlText(lines);
+    }
+
+    return finalizeTomlText(lines);
   }
 
   const normalizedUrl = trimmed.replace(/\s+/g, "");
   const replacementLine = `base_url = "${normalizedUrl}"`;
-  const pattern = /base_url\s*=\s*(["'])([^"']+)\1/;
 
-  if (pattern.test(normalizedText)) {
-    return normalizedText.replace(pattern, replacementLine);
+  if (targetSectionName) {
+    let targetSectionRange = getTomlSectionRange(lines, targetSectionName);
+    const targetMatch = targetSectionRange
+      ? findTomlAssignmentInRange(
+          lines,
+          TOML_BASE_URL_PATTERN,
+          targetSectionRange.bodyStartIndex,
+          targetSectionRange.bodyEndIndex,
+          targetSectionName,
+        )
+      : undefined;
+
+    if (targetMatch) {
+      lines[targetMatch.index] = replacementLine;
+      return finalizeTomlText(lines);
+    }
+
+    if (recoverableAssignments.length === 1) {
+      lines.splice(recoverableAssignments[0].index, 1);
+      targetSectionRange = getTomlSectionRange(lines, targetSectionName);
+    }
+
+    if (targetSectionRange) {
+      const insertIndex = getTomlSectionInsertIndex(lines, targetSectionRange);
+      lines.splice(insertIndex, 0, replacementLine);
+      return finalizeTomlText(lines);
+    }
+
+    if (lines.length > 0 && lines[lines.length - 1].trim() !== "") {
+      lines.push("");
+    }
+    lines.push(`[${targetSectionName}]`, replacementLine);
+    return finalizeTomlText(lines);
   }
 
-  const prefix =
-    normalizedText && !normalizedText.endsWith("\n")
-      ? `${normalizedText}\n`
-      : normalizedText;
-  return `${prefix}${replacementLine}\n`;
+  const topLevelEndIndex = getTopLevelEndIndex(lines);
+  const topLevelMatch = findTomlAssignmentInRange(
+    lines,
+    TOML_BASE_URL_PATTERN,
+    0,
+    topLevelEndIndex,
+  );
+  if (topLevelMatch) {
+    lines[topLevelMatch.index] = replacementLine;
+    return finalizeTomlText(lines);
+  }
+
+  const modelProviderIndex = getTopLevelModelProviderLineIndex(lines);
+  if (modelProviderIndex !== -1) {
+    lines.splice(modelProviderIndex + 1, 0, replacementLine);
+    return finalizeTomlText(lines);
+  }
+
+  if (lines.length === 0) {
+    return `${replacementLine}\n`;
+  }
+
+  const insertIndex = topLevelEndIndex;
+  lines.splice(insertIndex, 0, replacementLine);
+  return finalizeTomlText(lines);
 };
 
 // ========== Codex model name utils ==========
@@ -491,13 +779,16 @@ export const extractCodexModelName = (
 ): string | undefined => {
   try {
     const raw = typeof configText === "string" ? configText : "";
-    // 归一化中文/全角引号，避免正则提取失败
-    const text = normalizeQuotes(raw);
+    const text = normalizeTomlText(raw);
     if (!text) return undefined;
-
-    // 匹配 model = "xxx" 或 model = 'xxx'
-    const m = text.match(/^model\s*=\s*(['"])([^'"]+)\1/m);
-    return m && m[2] ? m[2] : undefined;
+    const lines = text.split("\n");
+    const topLevelMatch = findTomlAssignmentInRange(
+      lines,
+      TOML_MODEL_PATTERN,
+      0,
+      getTopLevelEndIndex(lines),
+    );
+    return topLevelMatch?.value;
   } catch {
     return undefined;
   }
@@ -509,47 +800,40 @@ export const setCodexModelName = (
   modelName: string,
 ): string => {
   const trimmed = modelName.trim();
-  // 归一化原文本中的引号（既能匹配，也能输出稳定格式）
-  const normalizedText = normalizeQuotes(configText);
+  const normalizedText = normalizeTomlText(configText);
+  const lines = normalizedText ? normalizedText.split("\n") : [];
+  const topLevelEndIndex = getTopLevelEndIndex(lines);
+  const topLevelMatch = findTomlAssignmentInRange(
+    lines,
+    TOML_MODEL_PATTERN,
+    0,
+    topLevelEndIndex,
+  );
 
-  // 允许清空：当 modelName 为空时，移除 model 行
   if (!trimmed) {
     if (!normalizedText) return normalizedText;
-    const next = normalizedText
-      .split("\n")
-      .filter((line) => !/^\s*model\s*=/.test(line))
-      .join("\n")
-      .replace(/\n{3,}/g, "\n\n")
-      .replace(/^\n+/, "");
-    return next;
+    if (topLevelMatch) {
+      lines.splice(topLevelMatch.index, 1);
+    }
+    return finalizeTomlText(lines);
   }
 
   const replacementLine = `model = "${trimmed}"`;
-  const pattern = /^model\s*=\s*["']([^"']+)["']/m;
-
-  if (pattern.test(normalizedText)) {
-    return normalizedText.replace(pattern, replacementLine);
+  if (topLevelMatch) {
+    lines[topLevelMatch.index] = replacementLine;
+    return finalizeTomlText(lines);
   }
 
-  // 如果不存在 model 字段，尝试在 model_provider 之后插入
-  // 如果 model_provider 也不存在，则插入到开头
-  const providerPattern = /^model_provider\s*=\s*["'][^"']+["']/m;
-  const match = normalizedText.match(providerPattern);
-
-  if (match && match.index !== undefined) {
-    // 在 model_provider 行之后插入
-    const endOfLine = normalizedText.indexOf("\n", match.index);
-    if (endOfLine !== -1) {
-      return (
-        normalizedText.slice(0, endOfLine + 1) +
-        replacementLine +
-        "\n" +
-        normalizedText.slice(endOfLine + 1)
-      );
-    }
+  const modelProviderIndex = getTopLevelModelProviderLineIndex(lines);
+  if (modelProviderIndex !== -1) {
+    lines.splice(modelProviderIndex + 1, 0, replacementLine);
+    return finalizeTomlText(lines);
   }
 
-  // 在文件开头插入
-  const lines = normalizedText.split("\n");
-  return `${replacementLine}\n${lines.join("\n")}`;
+  if (lines.length === 0) {
+    return `${replacementLine}\n`;
+  }
+
+  lines.splice(topLevelEndIndex, 0, replacementLine);
+  return finalizeTomlText(lines);
 };
diff --git a/tests/utils/providerConfigUtils.codex.test.ts b/tests/utils/providerConfigUtils.codex.test.ts
index 13dfa2ae..28e741ab 100644
--- a/tests/utils/providerConfigUtils.codex.test.ts
+++ b/tests/utils/providerConfigUtils.codex.test.ts
@@ -22,17 +22,21 @@ describe("Codex TOML utils", () => {
     expect(extractCodexModelName(output)).toBe("gpt-5-codex");
   });
 
-  it("removes model line when set to empty", () => {
+  it("removes only the top-level model line when set to empty", () => {
     const input = [
       'model_provider = "openai"',
       'base_url = "https://api.example.com/v1"',
       'model = "gpt-5-codex"',
       "",
+      "[profiles.default]",
+      'model = "profile-model"',
+      "",
     ].join("\n");
 
     const output = setCodexModelName(input, "");
 
-    expect(output).not.toMatch(/^\s*model\s*=/m);
+    expect(output).not.toMatch(/^model\s*=\s*"gpt-5-codex"$/m);
+    expect(output).toMatch(/^\[profiles\.default\]\nmodel = "profile-model"$/m);
     expect(extractCodexModelName(output)).toBeUndefined();
     expect(extractCodexBaseUrl(output)).toBe("https://api.example.com/v1");
   });
@@ -51,5 +55,97 @@ describe("Codex TOML utils", () => {
     const output2 = setCodexModelName(output1, " new-model \n");
     expect(extractCodexModelName(output2)).toBe("new-model");
   });
-});
 
+  it("reads and writes base_url in the active provider section", () => {
+    const input = [
+      'model_provider = "custom"',
+      'model = "gpt-5.4"',
+      "",
+      "[model_providers.custom]",
+      'name = "custom"',
+      'wire_api = "responses"',
+      "",
+      "[profiles.default]",
+      'approval_policy = "never"',
+      "",
+    ].join("\n");
+
+    const output = setCodexBaseUrl(input, "https://api.example.com/v1");
+
+    expect(output).toContain(
+      '[model_providers.custom]\nname = "custom"\nwire_api = "responses"\nbase_url = "https://api.example.com/v1"',
+    );
+    expect(extractCodexBaseUrl(output)).toBe("https://api.example.com/v1");
+  });
+
+  it("recovers a single misplaced base_url from another section", () => {
+    const input = [
+      'model_provider = "custom"',
+      'model = "gpt-5.4"',
+      "",
+      "[model_providers.custom]",
+      'name = "custom"',
+      'wire_api = "responses"',
+      "",
+      "[profiles.default]",
+      'approval_policy = "never"',
+      'base_url = "https://wrong.example/v1"',
+      "",
+    ].join("\n");
+
+    expect(extractCodexBaseUrl(input)).toBe("https://wrong.example/v1");
+
+    const output = setCodexBaseUrl(input, "https://fixed.example/v1");
+
+    expect(output).toContain(
+      '[model_providers.custom]\nname = "custom"\nwire_api = "responses"\nbase_url = "https://fixed.example/v1"',
+    );
+    expect(output).not.toContain("https://wrong.example/v1");
+    expect(output.match(/base_url\s*=/g)).toHaveLength(1);
+  });
+
+  it("does not treat mcp_servers base_url as provider base_url", () => {
+    const input = [
+      'model_provider = "azure"',
+      'model = "gpt-4"',
+      "",
+      "[model_providers.azure]",
+      'name = "Azure OpenAI"',
+      'wire_api = "responses"',
+      "",
+      "[mcp_servers.my_server]",
+      'base_url = "http://localhost:8080"',
+      "",
+    ].join("\n");
+
+    expect(extractCodexBaseUrl(input)).toBeUndefined();
+
+    const output = setCodexBaseUrl(input, "https://new.azure/v1");
+
+    expect(output).toContain(
+      '[model_providers.azure]\nname = "Azure OpenAI"\nwire_api = "responses"\nbase_url = "https://new.azure/v1"',
+    );
+    expect(output).toContain(
+      '[mcp_servers.my_server]\nbase_url = "http://localhost:8080"',
+    );
+  });
+
+  it("reads model only from the top-level config", () => {
+    const input = [
+      'model_provider = "custom"',
+      "",
+      "[profiles.default]",
+      'model = "profile-model"',
+      "",
+    ].join("\n");
+
+    expect(extractCodexModelName(input)).toBeUndefined();
+  });
+
+  it("handles single-quoted values", () => {
+    const input = "base_url = 'https://api.example.com/v1'\nmodel = 'gpt-5'\n";
+
+    expect(extractCodexBaseUrl(input)).toBe("https://api.example.com/v1");
+    expect(extractCodexModelName(input)).toBe("gpt-5");
+  });
+});