fix(mcp): allow third-party providers to approve project-scope .mcp.json servers (#696)

When a user runs openclaude with a third-party provider, project-scope
MCP servers added via `openclaude mcp add -s project ...` were silently
dropped from `/mcp` and `openclaude mcp list`. Re-adding the same
server printed "MCP server already exists in .mcp.json" but the server
never actually loaded. Reporter @gbmerrall pinpointed the cause to the
`if (usesAnthropicSetup)` gate around `handleMcpjsonServerApprovals`
in src/interactiveHelpers.tsx.

The MCP approval dialog and the CLAUDE.md external-includes warning
are about workspace trust, not about Anthropic auth. Gating them on
`usesAnthropicSetup` meant 3P-provider users never saw the dialog
that writes `enableAllProjectMcpServers: true` and
`enabledMcpjsonServers: [...]` to settings.local.json — without
those settings, the MCP server isn't loaded for use.

Drop the `usesAnthropicSetup` gate around the approval flow. The
inner logic is unchanged and `handleMcpjsonServerApprovals` already
early-returns when no servers are pending, so users without project
`.mcp.json` see no new behavior.

- src/interactiveHelpers.tsx: drop the gate, add a comment explaining
  why (and cite #696 so future readers can find context).
- src/__tests__/bugfixes.test.ts: +2 regression tests asserting the
  gate is gone and the issue is referenced.

Verified locally on Linux: build passes (v0.7.0), 1634 tests pass,
the 4 remaining failures (StartupScreen.test.ts, thinking.test.ts)
reproduce on main and are unrelated. The bundled dist/cli.mjs shows
`handleMcpjsonServerApprovals(root2)` running directly after
`setSessionTrustAccepted(true)` with no auth gate.

src/__tests__/bugfixes.test.ts

@@ -288,3 +288,30 @@ describe('Context overflow 500 fix', () => {
     expect(content).toContain('automatic compaction has failed')
   })
 })
+
+// ---------------------------------------------------------------------------
+// Fix N: Project-scope MCP servers from .mcp.json not detected for 3P providers (issue #696)
+// ---------------------------------------------------------------------------
+describe('Project-scope MCP approval — third-party providers (issue #696)', () => {
+  test('handleMcpjsonServerApprovals is NOT gated behind usesAnthropicSetup', async () => {
+    const content = await file('interactiveHelpers.tsx').text()
+
+    // The call site for handleMcpjsonServerApprovals must not sit inside an
+    // `if (usesAnthropicSetup) { ... }` block, or third-party providers will
+    // never get the dialog and project-scope .mcp.json servers will be silently
+    // dropped from /mcp listings (issue #696).
+    const approvalCallIdx = content.indexOf('await handleMcpjsonServerApprovals(root)')
+    expect(approvalCallIdx).toBeGreaterThan(-1)
+
+    // Look at the 800 chars BEFORE the call site for any `if (usesAnthropicSetup)`
+    // block that would still be open. Pick a window that's definitely inside the
+    // showSetupScreens function but not in earlier dialogs.
+    const before = content.slice(Math.max(0, approvalCallIdx - 800), approvalCallIdx)
+    expect(before).not.toMatch(/if\s*\(\s*usesAnthropicSetup\s*\)\s*{[^}]*$/)
+  })
+
+  test('issue #696 is referenced from the comment so future readers can find context', async () => {
+    const content = await file('interactiveHelpers.tsx').text()
+    expect(content).toContain('#696')
+  })
+})

src/interactiveHelpers.tsx

@@ -158,12 +158,14 @@ export async function showSetupScreens(root: Root, permissionMode: PermissionMod
     // Now that trust is established, prefetch system context if it wasn't already
     void getSystemContext();
 
-    // Skip MCP approval dialogs for third-party providers (no interactive auth prompts)
+    // MCP approval and external-includes warnings are about workspace
-    if (usesAnthropicSetup) {
+    // trust, not about Anthropic auth. They must run for all providers
-      // If settings are valid, check for any mcp.json servers that need approval
+    // — including third-party — otherwise project-scoped .mcp.json
-      const {
+    // servers never get the approval that writes
-        errors: allErrors
+    // enableAllProjectMcpServers / enabledMcpjsonServers into
-      } = getSettingsWithAllErrors();
+    // settings.local.json, and the servers are silently dropped from
+    // /mcp and `mcp list` (issue #696).
+    const { errors: allErrors } = getSettingsWithAllErrors();
     if (allErrors.length === 0) {
       await handleMcpjsonServerApprovals(root);
     }
@@ -177,7 +179,6 @@ export async function showSetupScreens(root: Root, permissionMode: PermissionMod
       await showSetupDialog(root, done => <ClaudeMdExternalIncludesDialog onDone={done} isStandaloneDialog externalIncludes={externalIncludes} />);
     }
   }
-  }
 
   // Track current repo path for teleport directory switching (fire-and-forget)
   // This must happen AFTER trust to prevent untrusted directories from poisoning the mapping

src/services/api/errors.openaiCompatibility.test.ts

@@ -28,38 +28,6 @@ test('maps endpoint_not_found category markers to actionable setup guidance', ()
   expect(text).toContain('/v1')
 })
 
-test('endpoint_not_found from a remote host shows the actual host, not Ollama (issue #926)', () => {
-  const error = APIError.generate(
-    404,
-    undefined,
-    'OpenAI API error 404: Not Found [openai_category=endpoint_not_found,host=integrate.api.nvidia.com] Hint: Endpoint at integrate.api.nvidia.com returned 404.',
-    new Headers(),
-  )
-
-  const message = getAssistantMessageFromError(error, 'moonshotai/kimi-k2.5-thinking')
-  const text = getFirstText(message)
-
-  expect(text).toContain('integrate.api.nvidia.com')
-  expect(text).toContain('moonshotai/kimi-k2.5-thinking')
-  expect(text).not.toContain('Ollama')
-  expect(text).not.toContain('11434')
-})
-
-test('endpoint_not_found without a host falls back to the Ollama-aware message', () => {
-  const error = APIError.generate(
-    404,
-    undefined,
-    'OpenAI API error 404: Not Found [openai_category=endpoint_not_found] Hint: Confirm OPENAI_BASE_URL includes /v1.',
-    new Headers(),
-  )
-
-  const message = getAssistantMessageFromError(error, 'qwen2.5-coder:7b')
-  const text = getFirstText(message)
-
-  expect(text).toContain('Provider endpoint was not found')
-  expect(text).toContain('Ollama')
-})
-
 test('maps tool_call_incompatible category markers to model/tool guidance', () => {
   const error = APIError.generate(
     400,

src/services/api/errors.ts

@@ -51,9 +51,7 @@ import {
 import { shouldProcessRateLimits } from '../rateLimitMocking.js' // Used for /mock-limits command
 import { extractConnectionErrorDetails, formatAPIError } from './errorUtils.js'
 import {
-  extractOpenAICategoryHost,
   extractOpenAICategoryMarker,
-  isLocalhostLikeHost,
   type OpenAICompatibilityFailureCategory,
 } from './openaiErrorClassification.js'
 
@@ -70,29 +68,25 @@ function mapOpenAICompatibilityFailureToAssistantMessage(options: {
   category: OpenAICompatibilityFailureCategory
   model: string
   rawMessage: string
-  host?: string
 }): AssistantMessage {
   const switchCmd = getIsNonInteractiveSession() ? '--model' : '/model'
   const compactHint = getIsNonInteractiveSession()
     ? 'Reduce prompt size or start a new session.'
     : 'Run /compact or start a new session with /new.'
-  const isLocalhost = options.host === undefined || isLocalhostLikeHost(options.host)
 
   switch (options.category) {
     case 'localhost_resolution_failed':
     case 'connection_refused':
       return createAssistantAPIErrorMessage({
-        content: isLocalhost
+        content:
-          ? 'Could not connect to the local OpenAI-compatible provider. Ensure the local server is running, then use OPENAI_BASE_URL=http://127.0.0.1:11434/v1 for Ollama.'
+          'Could not connect to the local OpenAI-compatible provider. Ensure the local server is running, then use OPENAI_BASE_URL=http://127.0.0.1:11434/v1 for Ollama.',
-          : `Could not connect to the provider at ${options.host}. Verify OPENAI_BASE_URL is correct and that the host is reachable.`,
         error: 'unknown',
       })
 
     case 'endpoint_not_found':
       return createAssistantAPIErrorMessage({
-        content: isLocalhost
+        content:
-          ? 'Provider endpoint was not found. Confirm OPENAI_BASE_URL targets an OpenAI-compatible /v1 endpoint (for Ollama: http://127.0.0.1:11434/v1).'
+          'Provider endpoint was not found. Confirm OPENAI_BASE_URL targets an OpenAI-compatible /v1 endpoint (for Ollama: http://127.0.0.1:11434/v1).',
-          : `Provider endpoint at ${options.host} returned 404. Verify OPENAI_BASE_URL is correct and that the selected model (${options.model}) is supported by this provider.`,
         error: 'invalid_request',
       })
 
@@ -573,7 +567,6 @@ export function getAssistantMessageFromError(
         category: openaiCategory,
         model,
         rawMessage: error.message,
-        host: extractOpenAICategoryHost(error.message),
       })
     }
   }

src/services/api/openaiErrorClassification.test.ts

@@ -4,10 +4,8 @@ import {
   buildOpenAICompatibilityErrorMessage,
   classifyOpenAIHttpFailure,
   classifyOpenAINetworkFailure,
-  extractOpenAICategoryHost,
   extractOpenAICategoryMarker,
   formatOpenAICategoryMarker,
-  isLocalhostLikeHost,
 } from './openaiErrorClassification.js'
 
 test('classifies localhost ECONNREFUSED as connection_refused', () => {
@@ -97,58 +95,3 @@ test('ignores unknown category markers during extraction', () => {
   const malformed = 'OpenAI API error 500 [openai_category=totally_fake_category]'
   expect(extractOpenAICategoryMarker(malformed)).toBeUndefined()
 })
-
-test('endpoint_not_found 404 from a remote host gets a host-aware hint (issue #926)', () => {
-  const failure = classifyOpenAIHttpFailure({
-    status: 404,
-    body: 'Not Found',
-    url: 'https://integrate.api.nvidia.com/v1/chat/completions',
-  })
-
-  expect(failure.category).toBe('endpoint_not_found')
-  expect(failure.requestUrl).toBe('https://integrate.api.nvidia.com/v1/chat/completions')
-  expect(failure.hint).toContain('integrate.api.nvidia.com')
-  expect(failure.hint).not.toContain('local providers')
-})
-
-test('endpoint_not_found 404 from localhost keeps the Ollama-flavored hint', () => {
-  const failure = classifyOpenAIHttpFailure({
-    status: 404,
-    body: 'Not Found',
-    url: 'http://127.0.0.1:11434/v1/chat/completions',
-  })
-
-  expect(failure.category).toBe('endpoint_not_found')
-  expect(failure.hint).toContain('local providers')
-})
-
-test('marker round-trip preserves host segment', () => {
-  const formatted = buildOpenAICompatibilityErrorMessage(
-    'OpenAI API error 404: Not Found',
-    {
-      category: 'endpoint_not_found',
-      hint: 'Endpoint at integrate.api.nvidia.com returned 404.',
-      requestUrl: 'https://integrate.api.nvidia.com/v1/chat/completions',
-    },
-  )
-
-  expect(formatted).toContain('[openai_category=endpoint_not_found,host=integrate.api.nvidia.com]')
-  expect(extractOpenAICategoryMarker(formatted)).toBe('endpoint_not_found')
-  expect(extractOpenAICategoryHost(formatted)).toBe('integrate.api.nvidia.com')
-})
-
-test('marker without host stays backward-compatible', () => {
-  const marker = formatOpenAICategoryMarker('endpoint_not_found')
-  expect(marker).toBe('[openai_category=endpoint_not_found]')
-  expect(extractOpenAICategoryMarker(marker)).toBe('endpoint_not_found')
-  expect(extractOpenAICategoryHost(marker)).toBeUndefined()
-})
-
-test('isLocalhostLikeHost matches loopback variants', () => {
-  expect(isLocalhostLikeHost('localhost')).toBe(true)
-  expect(isLocalhostLikeHost('127.0.0.1')).toBe(true)
-  expect(isLocalhostLikeHost('127.0.0.5')).toBe(true)
-  expect(isLocalhostLikeHost('::1')).toBe(true)
-  expect(isLocalhostLikeHost('integrate.api.nvidia.com')).toBe(false)
-  expect(isLocalhostLikeHost(undefined)).toBe(false)
-})

src/services/api/openaiErrorClassification.ts

@@ -21,7 +21,6 @@ export type OpenAICompatibilityFailure = {
   hint?: string
   code?: string
   status?: number
-  requestUrl?: string
 }
 
 const OPENAI_CATEGORY_MARKER_PREFIX = '[openai_category='
@@ -97,11 +96,6 @@ function isLocalhostLikeHostname(hostname: string | null): boolean {
   return /^127\./.test(hostname)
 }
 
-export function isLocalhostLikeHost(host: string | null | undefined): boolean {
-  if (!host) return false
-  return isLocalhostLikeHostname(host.toLowerCase())
-}
-
 function isContextOverflowMessage(body: string): boolean {
   const lower = body.toLowerCase()
   return (
@@ -155,18 +149,14 @@ function isModelNotFoundMessage(body: string): boolean {
 
 export function formatOpenAICategoryMarker(
   category: OpenAICompatibilityFailureCategory,
-  host?: string,
 ): string {
-  if (host && /^[A-Za-z0-9.\-:]+$/.test(host)) {
-    return `${OPENAI_CATEGORY_MARKER_PREFIX}${category},host=${host}]`
-  }
   return `${OPENAI_CATEGORY_MARKER_PREFIX}${category}]`
 }
 
 export function extractOpenAICategoryMarker(
   message: string,
 ): OpenAICompatibilityFailureCategory | undefined {
-  const match = message.match(/\[openai_category=([a-z_]+)(?:,host=[^\]]+)?]/)
+  const match = message.match(/\[openai_category=([a-z_]+)]/)
   const category = match?.[1]
 
   if (!category || !isOpenAICompatibilityFailureCategory(category)) {
@@ -176,17 +166,11 @@ export function extractOpenAICategoryMarker(
   return category
 }
 
-export function extractOpenAICategoryHost(message: string): string | undefined {
-  const match = message.match(/\[openai_category=[a-z_]+,host=([A-Za-z0-9.\-:]+)]/)
-  return match?.[1]
-}
-
 export function buildOpenAICompatibilityErrorMessage(
   baseMessage: string,
-  failure: Pick<OpenAICompatibilityFailure, 'category' | 'hint' | 'requestUrl'>,
+  failure: Pick<OpenAICompatibilityFailure, 'category' | 'hint'>,
 ): string {
-  const host = failure.requestUrl ? getHostname(failure.requestUrl) ?? undefined : undefined
+  const marker = formatOpenAICategoryMarker(failure.category)
-  const marker = formatOpenAICategoryMarker(failure.category, host)
   const hint = failure.hint ? ` Hint: ${failure.hint}` : ''
   return `${baseMessage} ${marker}${hint}`
 }
@@ -263,11 +247,8 @@ export function classifyOpenAINetworkFailure(
 export function classifyOpenAIHttpFailure(options: {
   status: number
   body: string
-  url?: string
 }): OpenAICompatibilityFailure {
   const body = options.body ?? ''
-  const hostname = options.url ? getHostname(options.url) : null
-  const isLocalHost = isLocalhostLikeHostname(hostname)
 
   if (options.status === 401 || options.status === 403) {
     return {
@@ -303,17 +284,13 @@ export function classifyOpenAIHttpFailure(options: {
   }
 
   if (options.status === 404) {
-    const isRemote = hostname !== null && !isLocalHost
     return {
       source: 'http',
       category: 'endpoint_not_found',
       retryable: false,
       status: options.status,
       message: body,
-      requestUrl: options.url,
+      hint: 'Endpoint was not found. Confirm OPENAI_BASE_URL includes /v1 for OpenAI-compatible local providers.',
-      hint: isRemote
-        ? `Endpoint at ${hostname} returned 404. Verify OPENAI_BASE_URL is correct and the requested model is supported by this provider.`
-        : 'Endpoint was not found. Confirm OPENAI_BASE_URL includes /v1 for OpenAI-compatible local providers.',
     }
   }
 

src/services/api/openaiShim.ts

@@ -1935,9 +1935,7 @@ class OpenAIShimMessages {
         classifyOpenAIHttpFailure({
           status,
           body: errorBody,
-          url: requestUrl,
         })
-      const failureWithUrl = { ...failure, requestUrl: failure.requestUrl ?? requestUrl }
       const redactedUrl = redactUrlForDiagnostics(requestUrl)
 
       logForDebugging(
@@ -1950,7 +1948,7 @@ class OpenAIShimMessages {
         parsedBody,
         buildOpenAICompatibilityErrorMessage(
           `OpenAI API error ${status}: ${errorBody}${rateHint}`,
-          failureWithUrl,
+          failure,
         ),
         responseHeaders,
       )