MCP server tools are not exposed to the model when using the built-in lmstudio provider (identical setup works with the openai provider)
What version of Codex CLI is running?
codex-cli 0.139.0
What subscription do you have?
ChatGPT Enterprise — but the bug concerns the built-in lmstudio provider with a local model, where no subscription is involved; the subscription account was only used for the working gpt-5.5 control case.
Which model were you using?
google/gemma-4-26b-a4b-qat vs. gpt-5.5
What platform is your computer?
Microsoft Windows NT 10.0.26200.0 x64
What terminal emulator and version are you using (if applicable)?
Windows Terminal — PowerShell 7.6.2; no multiplexer.
Codex doctor report
{
"schemaVersion": 1,
"generatedAt": "1781171963s since unix epoch",
"overallStatus": "warning",
"codexVersion": "0.139.0",
"checks": {
"app_server.status": {
"id": "app_server.status",
"category": "app-server",
"status": "ok",
"summary": "background server is not running",
"details": {
"control socket": "C:\\Users\\MatthiasKlöpper\\.codex\\app-server-control\\app-server-control.sock",
"daemon state dir": "C:\\Users\\MatthiasKlöpper\\.codex\\app-server-daemon",
"mode": "ephemeral",
"pid file": "C:\\Users\\MatthiasKlöpper\\.codex\\app-server-daemon\\app-server.pid (missing)",
"settings": "C:\\Users\\MatthiasKlöpper\\.codex\\app-server-daemon\\settings.json (missing)",
"status": "not running",
"update-loop pid file": "C:\\Users\\MatthiasKlöpper\\.codex\\app-server-daemon\\app-server-updater.pid (missing)"
},
"remediation": null,
"durationMs": 0
},
"auth.credentials": {
"id": "auth.credentials",
"category": "auth",
"status": "ok",
"summary": "auth is configured",
"details": {
"auth file": "C:\\Users\\MatthiasKlöpper\\.codex\\auth.json",
"auth storage mode": "File",
"stored API key": "false",
"stored ChatGPT tokens": "true",
"stored agent identity": "false",
"stored auth mode": "chatgpt"
},
"remediation": null,
"durationMs": 0
},
"config.load": {
"id": "config.load",
"category": "config",
"status": "ok",
"summary": "config loaded",
"details": {
"CODEX_HOME": "C:\\Users\\MatthiasKlöpper\\.codex",
"config.toml": "C:\\Users\\MatthiasKlöpper\\.codex\\config.toml",
"config.toml parse": "ok",
"cwd": "C:\\Users\\MatthiasKlöpper",
"enabled feature flags": "shell_tool, shell_snapshot, terminal_resize_reflow, sqlite, hooks, enable_request_compression, multi_agent, apps, tool_suggest, plugins, in_app_browser, browser_use, browser_use_external, computer_use, plugin_sharing, image_generation, skill_mcp_dependency_install, steer, guardian_approval, goals, collaboration_modes, tool_call_mcp_elicitation, personality, fast_mode, tui_app_server, workspace_dependencies",
"feature flag overrides": "none",
"feature flags enabled": "26",
"log dir": "C:\\Users\\MatthiasKlöpper\\.codex\\log",
"mcp servers": "1",
"model": "gpt-5.5",
"model provider": "openai",
"sqlite home": "C:\\Users\\MatthiasKlöpper\\.codex"
},
"remediation": null,
"durationMs": 0
},
"git.environment": {
"id": "git.environment",
"category": "git",
"status": "ok",
"summary": "git version 2.53.0.windows.3",
"details": {
"PATH git #1": "C:\\Program Files\\Git\\cmd\\git.exe",
"PATH git entries": "1",
"git build options": "git version 2.53.0.windows.3; cpu: x86_64; built from commit: f8165afd89b0c190677a093f20894f5fce12f97a; sizeof-long: 4; sizeof-size_t: 8; shell-path: D:/git-sdk-64/usr/bin/sh; rust: disabled; feature: fsmonitor--daemon; gettext: enabled; libcurl: 8.18.0; OpenSSL: OpenSSL 3.5.5 27 Jan 2026; zlib: 1.3.1; SHA-1: SHA1_DC; SHA-256: SHA256_BLK; default-ref-format: files; default-hash: sha1",
"git exec path": "C:/Program Files/Git/mingw64/libexec/git-core",
"git version": "git version 2.53.0.windows.3",
"repo detected": "false",
"selected git": "C:\\Program Files\\Git\\cmd\\git.exe"
},
"remediation": null,
"durationMs": 53
},
"installation": {
"id": "installation",
"category": "install",
"status": "ok",
"summary": "installation looks consistent",
"details": {
"PATH codex #1": "C:\\Users\\MatthiasKl�pper\\AppData\\Local\\Programs\\OpenAI\\Codex\\bin\\codex.exe",
"PATH codex #2": "C:\\Users\\MatthiasKl�pper\\AppData\\Roaming\\npm\\codex",
"PATH codex #3": "C:\\Users\\MatthiasKl�pper\\AppData\\Roaming\\npm\\codex.cmd",
"PATH codex entries": "3",
"current executable": "C:\\Users\\MatthiasKlöpper\\AppData\\Local\\Programs\\OpenAI\\Codex\\bin\\codex.exe",
"install context": "standalone (windows, package C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc, bin C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc\\bin, resources C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc\\codex-resources, path C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc\\codex-path)",
"managed by bun": "false",
"managed by npm": "false",
"managed package root": "not set"
},
"remediation": null,
"durationMs": 41
},
"mcp.config": {
"id": "mcp.config",
"category": "mcp",
"status": "ok",
"summary": "MCP configuration is locally consistent",
"details": {
"configured servers": "1",
"disabled servers": "0",
"stdio servers": "1"
},
"remediation": null,
"durationMs": 5
},
"network.env": {
"id": "network.env",
"category": "network",
"status": "ok",
"summary": "network-related environment looks readable",
"details": {
"proxy env vars": "none"
},
"remediation": null,
"durationMs": 0
},
"network.provider_reachability": {
"id": "network.provider_reachability",
"category": "reachability",
"status": "ok",
"summary": "active provider endpoints are reachable over HTTP",
"details": {
"ChatGPT base URL": "https://chatgpt.com/backend-api/ reachable (HTTP 403)",
"reachability mode": "ChatGPT auth"
},
"remediation": null,
"durationMs": 91
},
"network.websocket_reachability": {
"id": "network.websocket_reachability",
"category": "websocket",
"status": "ok",
"summary": "Responses WebSocket handshake succeeded",
"details": {
"DNS": "2 IPv4, 2 IPv6, first IPv6",
"auth mode": "chatgpt",
"connect timeout": "15000 ms",
"endpoint": "wss://chatgpt.com/backend-api/<redacted>",
"handshake result": "HTTP 101 Switching Protocols",
"model provider": "openai",
"models etag present": "true",
"provider name": "OpenAI",
"proxy env vars": "none",
"reasoning header": "false",
"server model present": "false",
"supports websockets": "true",
"wire API": "responses"
},
"remediation": null,
"durationMs": 849
},
"runtime.provenance": {
"id": "runtime.provenance",
"category": "runtime",
"status": "ok",
"summary": "running standalone on windows-x86_64",
"details": {
"commit": "unknown",
"current executable": "C:\\Users\\MatthiasKlöpper\\AppData\\Local\\Programs\\OpenAI\\Codex\\bin\\codex.exe",
"install method": "standalone (windows, package C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc, bin C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc\\bin, resources C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc\\codex-resources, path C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc\\codex-path)",
"platform": "windows-x86_64",
"version": "0.139.0"
},
"remediation": null,
"durationMs": 0
},
"runtime.search": {
"id": "runtime.search",
"category": "search",
"status": "ok",
"summary": "search is OK (bundled)",
"details": {
"search command": "C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases\\0.139.0-x86_64-pc-windows-msvc\\codex-path\\rg.exe",
"search command readiness": "file exists",
"search provider": "bundled"
},
"remediation": null,
"durationMs": 0
},
"sandbox.helpers": {
"id": "sandbox.helpers",
"category": "sandbox",
"status": "ok",
"summary": "sandbox configuration is readable",
"details": {
"approval policy": "OnRequest",
"codex-linux-sandbox helper": "none",
"execve wrapper helper": "none",
"filesystem sandbox": "unrestricted",
"network sandbox": "enabled"
},
"remediation": null,
"durationMs": 0
},
"state.paths": {
"id": "state.paths",
"category": "state",
"status": "ok",
"summary": "state paths and databases are inspectable",
"details": {
"CODEX_HOME": "C:\\Users\\MatthiasKlöpper\\.codex (dir)",
"active rollout files": "13 files, 1911759 total bytes, 147058 average bytes",
"archived rollout files": "18 files, 1681555 total bytes, 93419 average bytes",
"goals DB": "C:\\Users\\MatthiasKlöpper\\.codex\\goals_1.sqlite (file)",
"goals DB integrity": "ok",
"log DB": "C:\\Users\\MatthiasKlöpper\\.codex\\logs_2.sqlite (file)",
"log DB integrity": "ok",
"log dir": "C:\\Users\\MatthiasKlöpper\\.codex\\log (missing)",
"memories DB": "C:\\Users\\MatthiasKlöpper\\.codex\\memories_1.sqlite (file)",
"memories DB integrity": "ok",
"sqlite home": "C:\\Users\\MatthiasKlöpper\\.codex (dir)",
"standalone release cache": "1 entries in C:\\Users\\MatthiasKlöpper\\.codex\\packages\\standalone\\releases",
"state DB": "C:\\Users\\MatthiasKlöpper\\.codex\\state_5.sqlite (file)",
"state DB integrity": "ok"
},
"remediation": null,
"durationMs": 344
},
"state.rollout_db_parity": {
"id": "state.rollout_db_parity",
"category": "threads",
"status": "warning",
"summary": "rollout files and state DB thread inventory differ",
"details": {
"default model provider": "openai",
"rollout DB active files": "13",
"rollout DB active rows": "13",
"rollout DB archive mismatches": "0",
"rollout DB archived files": "18",
"rollout DB archived rows": "18",
"rollout DB duplicate DB paths": "0",
"rollout DB duplicate rollout thread ids": "0",
"rollout DB malformed file names": "0",
"rollout DB missing active rows": "0",
"rollout DB missing archived rows": "1",
"rollout DB missing archived sample": "C:\\Users\\MatthiasKlöpper\\.codex\\archived_sessions\\rollout-2026-06-11T03-41-08-019eb457-1e7b-7973-b885-95b4a0a5f550.jsonl",
"rollout DB model providers": "lmstudio=24, openai=7",
"rollout DB rows": "31",
"rollout DB scan cap reached": "false",
"rollout DB scan errors": "0",
"rollout DB sources": "vscode=21, cli=10",
"rollout DB stale row sample": "C:\\Users\\MatthiasKlöpper\\.codex\\sessions\\2026\\06\\11\\rollout-2026-06-11T03-41-08-019eb457-1e7b-7973-b885-95b4a0a5f550.jsonl",
"rollout DB stale rows": "1"
},
"issues": [
{
"severity": "warning",
"cause": "rollout files are missing from the state DB",
"measured": "0 active, 1 archived",
"expected": "every rollout file has a matching threads row",
"remedy": null,
"fields": []
},
{
"severity": "warning",
"cause": "state DB rows point at missing or unusable rollout files",
"measured": "1 stale rows",
"expected": "every state DB rollout path is a file on disk",
"remedy": null,
"fields": []
}
],
"remediation": null,
"durationMs": 713
},
"system.environment": {
"id": "system.environment",
"category": "system",
"status": "ok",
"summary": "OS language de-DE",
"details": {
"EDITOR": "not set",
"VISUAL": "not set",
"os": "Windows 10.0.26200 (Windows 11 Enterprise) [64-bit]",
"os language": "de-DE",
"os type": "Windows",
"os version": "10.0.26200"
},
"remediation": null,
"durationMs": 0
},
"terminal.env": {
"id": "terminal.env",
"category": "terminal",
"status": "ok",
"summary": "terminal metadata was detected",
"details": {
"color output": "enabled",
"console input code page": "850",
"console output code page": "850",
"stderr console mode": "0x00000007 (VT processing: true)",
"stderr is terminal": "true",
"stdin is terminal": "true",
"stdout console mode": "0x00000007 (VT processing: true)",
"stdout is terminal": "true",
"terminal": "unknown",
"terminal size": "120x30"
},
"remediation": null,
"durationMs": 0
},
"terminal.title": {
"id": "terminal.title",
"category": "title",
"status": "ok",
"summary": "terminal title default",
"details": {
"terminal title activity": "true",
"terminal title items": "activity, project-name",
"terminal title project source": "cwd",
"terminal title project value": "MatthiasKlöpper",
"terminal title source": "default"
},
"remediation": null,
"durationMs": 0
},
"updates.status": {
"id": "updates.status",
"category": "updates",
"status": "ok",
"summary": "update configuration is locally consistent",
"details": {
"cached latest version": "0.139.0",
"check for update on startup": "true",
"last checked at": "2026-06-10T23:45:37.180971300Z",
"latest version": "0.139.0",
"latest version status": "current version is not older",
"update action": "standalone installer",
"version cache": "C:\\Users\\MatthiasKlöpper\\.codex\\version.json"
},
"remediation": null,
"durationMs": 229
}
}
}
What issue are you seeing?
MCP server tools are not exposed to the model when using the built-in lmstudio provider (identical setup works with the openai provider)
Versions / platform
- codex-cli: 0.139.0 (npm install and Codex desktop app bundled CLI, both same version)
- OS: Windows 11 x64
- Local backend: LM Studio 0.4.15,
/v1/responsesendpoint, modelgoogle/gemma-4-26b-a4b-qat(GGUF) - Provider: built-in
lmstudioprovider (no custom[model_providers]block)
Summary
An MCP server configured under [mcp_servers.playwright] starts correctly and its tools are registered with the harness (visible via /mcp). When the session runs against the built-in lmstudio provider, none of the MCP tools are included in the tool list delivered to the model — the model only receives Codex built-ins (shell_command, apply_patch, update_plan, ...). Switching only the model to gpt-5.5 (openai provider), same config and same MCP server, delivers the full MCP toolset to the model.
This looks like the provider-specific branch of session tool-list serialization dropping MCP entries for local providers, related to but distinct from #19871 (which reports unreliable invocation; here the tools are absent from the model's tool list entirely). The general case in #20771 (tools invisible on all providers, v0.128.0) does not reproduce on 0.139.0 — the openai provider path works.
Configuration (relevant excerpt)
model = "google/gemma-4-26b-a4b-qat"
model_provider = "lmstudio"
[mcp_servers.playwright]
enabled = true
command = "npx"
args = ["-y", "@playwright/mcp@latest", "--browser", "msedge", "--image-responses", "omit"]
startup_timeout_sec = 120
tool_timeout_sec = 600
Reproduction
- Start LM Studio with any tool-capable model loaded and the local server on port 1234.
- Use the config above (built-in
lmstudioprovider, any stdio MCP server). - Launch
codex, run/mcp— the playwright server is connected and lists ~24browser_*tools. ✅ - Prompt:
List the exact names of every tool/function you can call, verbatim, one per line. Do not execute anything. - The model lists only built-ins:
shell_command,list_mcp_resources,list_mcp_resource_templates,read_mcp_resource,update_plan,request_user_input,view_image,get_goal,create_goal,update_goal,apply_patch. Nomcp__playwright.*entries. ❌ - Direct invocation (
Call mcp__playwright.browser_navigate with url https://example.com) fails; the model reports the tool does not exist and falls back to shell improvisation. - Re-run steps 4–6 with
codex --model gpt-5.5(openai provider), same config, same MCP server process: the verbatim tool list now contains the full set, e.g.mcp__playwright.browser_navigate,mcp__playwright.browser_snapshot, ... (24 tools), plus connector tools (mcp__codex_apps__github.*). Tool calls work. ✅
Evidence
/mcp output (lmstudio session — harness side is healthy):
• playwright
• Auth: Unsupported
• Tools: browser_click, browser_close, browser_console_messages, browser_drag,
browser_drop, browser_evaluate, browser_file_upload, browser_fill_form,
browser_handle_dialog, browser_hover, browser_navigate, browser_navigate_back,
browser_network_request, browser_network_requests, browser_press_key,
browser_resize, browser_run_code_unsafe, browser_select_option, browser_snapshot,
browser_tabs, browser_take_screenshot, browser_type, browser_wait_for
Model's verbatim tool list, lmstudio provider (gemma-4-26b-a4b-qat):
shell_command
list_mcp_resources
list_mcp_resource_templates
read_mcp_resource
update_plan
request_user_input
view_image
get_goal
create_goal
update_goal
apply_patch
Model's verbatim tool list, openai provider (gpt-5.5), same config — excerpt:
functions.shell_command
...
mcp__playwright.browser_click
mcp__playwright.browser_close
...
mcp__playwright.browser_navigate
mcp__playwright.browser_snapshot
...
Expected behavior
MCP server tools should be included in the tool list sent to the model regardless of provider, matching what /mcp reports.
Actual behavior
For the built-in lmstudio provider, MCP tools are silently omitted from the model-visible tool list. Built-in tools are delivered and callable, so the tools array itself does reach the model.
Additional observations
- A warning is emitted at session start:
Model metadata for 'google/gemma-4-26b-a4b-qat' not found. Defaulting to fallback metadata; this can degrade performance and cause issues.Possibly relevant if fallback metadata gates MCP tool exposure. - Reproduces identically in the Codex desktop app (Windows) and the CLI.
- The MCP server itself is healthy: it starts within the timeout, responds to
tools/list, and works end-to-end in the gpt-5.5 session.
Related issues
- #19871 — MCP tool invocation regressed for custom/local providers (Responses API) in v0.117.0+
- #20771 — MCP servers load but tools not exposed to the model (closed; that general case does not reproduce on 0.139.0 — the openai provider path works here)
What steps can reproduce the bug?
- Start LM Studio with any tool-capable model loaded and the local server on port 1234.
- Use the config above (built-in
lmstudioprovider, any stdio MCP server). - Launch
codex, run/mcp— the playwright server is connected and lists ~24browser_*tools. ✅ - Prompt:
List the exact names of every tool/function you can call, verbatim, one per line. Do not execute anything. - The model lists only built-ins:
shell_command,list_mcp_resources,list_mcp_resource_templates,read_mcp_resource,update_plan,request_user_input,view_image,get_goal,create_goal,update_goal,apply_patch. Nomcp__playwright.*entries. ❌ - Direct invocation (
Call mcp__playwright.browser_navigate with url https://example.com) fails; the model reports the tool does not exist and falls back to shell improvisation. - Re-run steps 4–6 with
codex --model gpt-5.5(openai provider), same config, same MCP server process: the verbatim tool list now contains the full set, e.g.mcp__playwright.browser_navigate,mcp__playwright.browser_snapshot, ... (24 tools), plus connector tools (mcp__codex_apps__github.*). Tool calls work. ✅
What is the expected behavior?
MCP server tools should be included in the tool list sent to the model regardless of provider, matching what /mcp reports.
Additional information
_No response_
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗