M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
A

Ask LLM MCP Server

Community
Give your AI coding assistant a second opinion from a different model
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #133 of 438
Stars
★ 18 Very popular
Transport
stdio (local process)
Runtime
Node.js 20+
Credentials
Optional API key
License
MIT
Last commit
Tools
14
59FMRS · C

Ask LLM is a third-party, MIT-licensed MCP server maintained by Lykhoyda that offers unified multi-model orchestration across Codex, Claude, Grok, Antigravity, Ollama, and Gemini, plus auxiliary tools such as multi-llm fan-out, diagnostics, usage stats, and sessions. It suits developers who already have provider CLIs installed and authenticated and want code review or second opinions, but it requires the user to handle key configuration, data-transfer exposure, and metered billing risks; note the Gemini enterprise-seat restriction and Antigravity's experimental status. The project is unofficial and unaffiliated with the model vendors.

Strongest · Documentation 14/20 Weakest · Reliability 9/20

Reliability
9/20
Security and permissions
12/20
Maintenance
11/20
Documentation
14/20
Setup experience
13/20
Why each score
Reliability 9/20
Static evidence is limited to the README and the server.json manifest, with no source files, committed tests, or CI workflow contents. The manifest is structurally complete (name, version 0.2.0, npm identifier @ask-llm/mcp, stdio transport, 16 environment variables with defaults) and the declared tools are consistent with the README. However, the MCP init handshake, tool schemas, and runtime behavior cannot be verified; coupling to six-plus external provider CLIs, 10-minute default timeouts, and fallback chains (Gemini/Codex quota, Claude overload) are asserted only in prose. Per static calibration, reliability is capped at 12; with key paths unverifiable and error handling described only textually, the score is 9.
Security and permissions 12/20
Security posture is relatively articulated in the manifest: secret-typed variables (XAI_API_KEY, CURSOR_API_KEY) are marked isSecret: true; providers default to advisory read-only modes (Claude limited to Read/Glob/Grep, ask-cursor-agent read-only ask mode, Antigravity sandboxed by default), and the manifest claims no billing/credit enablement, no Grok harness or model fallback, and a 16384-token output ceiling to bound spend. Residual risks: prompts and repository content are sent to third-party model vendors (external network data boundary) without a manifest-level confirmation or disclosure mechanism; sessionId persistence, Ollama replay, and Cursor/agy CLI invocation scope lack source-level verification; the status of 11 open issues is unknown. No red-line evidence (malware, credential theft, covert exfiltration, irreversible defaults, real tokens in examples) was found, so the 0-4 penalty is not triggered; permissions/confirmation and data-flow disclosure remain incomplete, hence 12.
Maintenance 11/20
MIT license, version 0.2.0 consistent between manifest and README, plus CI and release badges, a package migration table, a contributing guide, and a documentation site all indicate active maintenance intent and a release process. Deductions: only 18 stars; 11 open issues with no evidence of response latency or triage; dependency update strategy for seven provider CLIs and multiple npm packages is absent; no SECURITY.md or vulnerability-disclosure channel is present in the supplied material; no commit history, release cadence, or independently verifiable maintainer identity. Static evidence does not support a high score, so 11.
Documentation 14/20
Documentation is layered: quick start for Claude Code, Cursor, Codex CLI, Claude Desktop, Pi, and generic STDIO; a provider selection table with default/fallback models and prerequisites (Codex account, Claude Code CLI, XAI_API_KEY, AI Pro/Ultra, local Ollama, enterprise Gemini); explicit disclosure that Gemini CLI became enterprise-gated on 2026-06-18 with alternatives; tool parameter notes (sessionId, outputSchema, usage://current-session resource); doctor diagnostics with --json/--format; cost and timeout defaults; a migration mapping; a disclaimer; and llms.txt for agents. Deductions: model names (gpt-6-astra, grok-4.7, qwen3.8:27b) cannot be corroborated from the supplied material; no per-error-code troubleshooting; some behaviors (codex-pair marker authorization, Fable verification limits) are disclosed but ambiguous; all documentation is repository self-report and is treated as untrusted evidence only. Score 14.
Setup experience 13/20
The install path is clear and short: one command (npx -y @ask-llm/mcp) for Claude Code, plus explicit config snippets for Cursor, Codex CLI, Claude Desktop, and generic STDIO clients; doctor diagnostics, a REPL, and a globally installed ask-llm-mcp binary are offered as alternatives. Deductions: prerequisites require Node.js 20+ and Linux/macOS only, with Windows unaddressed; real usability depends on the user installing and authenticating multiple external provider CLIs (Codex, Claude Code, agy >= 1.1.5, gemini login, local Ollama models); Ollama requires a manual pull and Gemini is enterprise-gated, so per-provider failure degrades the corresponding tools; 16 environment variables, though all optional, add configuration complexity; no verifiable installation evidence. Static setup cap is 15, so 13.

Static review · not runReviewed 2026-09-25

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesRuns commands or codeUses the network

Best for

  • Developers who have at least one supported provider CLI installed and authenticated, or the relevant API keys configured
  • Teams that want cross-model code review and plan debate inside their coding workflow
  • Users who need private or offline review via Ollama
  • Users of MCP clients such as Claude Code, Codex CLI, Cursor, and Claude Desktop

Not for

  • Users without any provider CLI installed or keys configured (the server does not supply model access itself)
  • Environments that cannot install Node.js 20+ or are not Linux/macOS
  • Users expecting a fully hosted, zero-config, or officially supported product
  • Scenarios where outbound data transfer is strictly forbidden and no local provider such as Ollama is available

Required permissions

  • Read local workspace files for context (e.g. Claude provider's Read/Glob/Grep-only access, Gemini's @ file syntax)
  • Launch and call installed provider CLIs locally (codex, claude, agy, gemini, agent, etc.)
  • Network access to call the xAI API (when using XAI_API_KEY) and the local Ollama service (default http://localhost:11434)
  • Read API keys from environment variables (e.g. XAI_API_KEY, CURSOR_API_KEY)
  • Cursor Agent runs in read-only ask mode and does not change force/trust/spend settings

Risks and side effects

  • Code and prompts may be sent to external model providers, creating data-transfer exposure; choose Ollama when local privacy is required
  • Metered endpoints such as the xAI API incur real costs; ASK_GROK_MAX_OUTPUT_TOKENS only bounds spend, it does not make it free
  • Grok and Ollama never fall back to another model, so requests may fail outright
  • Gemini CLI has been enterprise-seat-only since 2026-06-18; non-enterprise accounts get guidance instead of output
  • Antigravity is experimental and one-shot, with limited reliability
  • Plugin workflows such as codex-pair can trigger repeated external calls and costs, requiring an explicit marker and consent
  • This is a third-party, unofficial tool with no affiliation with or endorsement by Anthropic, Google, OpenAI, or xAI

Setup

Before you start

Runtime：Node.js 20+

XAI_API_KEY optionalsecret xAI API key for metered Grok requests, obtained from the xAI console; optional if you use other providers.
CURSOR_API_KEY optionalsecret Optional Cursor CLI credential for ask-cursor-agent, an alternative to agent login.
Other optional settings (14)
ASK_GROK_HARNESS optional Which Grok harness to use: xai-api (default) or grok-cli; there is no failover between them.
ASK_GROK_MODEL optional Exact model override for the selected harness (API and CLI both default to grok-4.7, no fallback).
ASK_GROK_REASONING_EFFORT optional Grok reasoning effort: low, medium, high, or xhigh (default high).
ASK_GROK_MAX_OUTPUT_TOKENS optional xAI API output-token ceiling to bound accidental spend, default 16384.
ASK_GROK_TIMEOUT_MS optional Timeout for xAI API requests in milliseconds, default 600000.
ASK_CURSOR_TIMEOUT_MS optional Timeout for Cursor Agent harness calls in milliseconds, default 600000.
ASK_CLAUDE_MODEL optional Default Claude model alias or full model name, default opus.
ASK_CLAUDE_FALLBACK_MODEL optional Claude CLI fallback model used when Opus is overloaded or unavailable, default sonnet.
ASK_CLAUDE_TIMEOUT_MS optional Timeout for Claude CLI execution in milliseconds, default 600000 (10 minutes).
OLLAMA_HOST optional Ollama server address, default http://localhost:11434.
ASK_ANTIGRAVITY_TIMEOUT_MS optional Timeout for Antigravity (agy) execution in milliseconds, default 300000 (5 minutes).
ASK_ANTIGRAVITY_SANDBOX optional Set to 0 to drop agy's --sandbox flag if it blocks --add-dir context reads; sandbox is on by default.
GMCPT_TIMEOUT_MS optional Timeout for CLI execution in milliseconds, default 300000 (5 minutes).
GMCPT_LOG_LEVEL optional Log verbosity: debug, info, warn, or error (default warn).
  1. Ensure Node.js 20+ (Linux or macOS) and at least one provider CLI installed and authenticated.
  2. Claude Code: run claude mcp add --scope user ask-llm -- npx -y @ask-llm/mcp (or ask-llm-mcp after a global install of @ask-llm/mcp).
  3. Cursor: add {"mcpServers":{"ask-llm":{"command":"npx","args":["-y","@ask-llm/mcp"]}}} to .cursor/mcp.json or ~/.cursor/mcp.json.
  4. Codex CLI: add [mcp_servers.ask-llm] command = "npx", args = ["-y", "@ask-llm/mcp"] to ~/.codex/config.toml.
  5. Claude Desktop: add the same mcpServers entry to claude_desktop_config.json.
  6. Other STDIO clients: use {"command":"npx","args":["-y","@ask-llm/mcp"]}.
  7. Optionally set environment variables such as XAI_API_KEY or CURSOR_API_KEY, then run npx @ask-llm/mcp doctor to check the setup.
.mcp.json 
{"mcpServers":{"ask-llm":{"command":"npx","args":["-y","@ask-llm/mcp"]}}}

Shown for Claude Code. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "ask-llm": {
      "command": "npx",
      "args": [
        "-y",
        "@ask-llm/mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add ask-llm -- npx -y @ask-llm/mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After installing, the client's tool list should show ask-llm, multi-llm, ping and the other ask-* tools; sending "ask codex to review my last commit" and getting Codex's review back proves the connection. Run npx @ask-llm/mcp doctor to check Node version, PATH resolution and provider CLI presence.

Troubleshooting

  1. Run npx @ask-llm/mcp doctor to check Node version, PATH resolution, and provider CLI presence and versions
  2. Confirm the target provider CLI is installed and signed in, or that the required API key is passed via environment variables
  3. If a Codex session cannot be resumed, verify the first call passed sessionId as an empty string
  4. If Grok requests fail, confirm ASK_GROK_HARNESS matches the harness actually used, and note there is no failover
  5. If Ollama errors, follow its prompt to ollama pull the required model and verify OLLAMA_HOST
  6. If Antigravity context reads are blocked by the sandbox, set ASK_ANTIGRAVITY_SANDBOX=0 to drop the --sandbox flag
  7. If Gemini returns no output, confirm the account has an enterprise Gemini Code Assist seat; subscription users can switch to Antigravity
  8. Adjust ASK_*_TIMEOUT_MS values for timeouts (some default to 600000 ms, others to 300000 ms)

Things to try

Once connected, you can ask your AI assistant things like:

  • ask codex to review src/auth.ts for security issues
  • ask claude for an independent opinion on this architecture
  • use multi-llm to compare what codex and grok think about this approach
  • ask ollama to explain src/config.ts

Tools 14

ask-llm read-only
Unified orchestrator: pick a single provider per call, or fan out to every installed provider
multi-llm read-only
Send one prompt to multiple providers in parallel; returns per-provider responses and usage in one call
ask-codex read-only
Ask via Codex CLI; GPT-6 Astra with Terra fallback. Omit sessionId for ephemeral use, or pass sessionId as an empty string first to persist and resume
ask-claude read-only
Ask via Claude Code CLI; opus (Opus 5.5) with Sonnet fallback; native sessions; Read/Glob/Grep-only workspace access
ask-grok read-only
One-shot Grok prompt through an explicit xai-api (default) or grok-cli harness; exact harness model ID; no harness or model fallback
ask-cursor-agent read-only
Model-neutral Cursor Agent harness: choose a provider (claude, codex, gemini, grok) plus an exact Cursor catalog model verified against that family; read-only ask mode; no force/trust/spend changes and no fallback
ask-antigravity read-only
Google Antigravity (agy) for a subscription-backed second opinion. Experimental; one-shot
ask-ollama read-only
Local Ollama; fully private, zero cost; server-side conversation replay via sessionId
Show 6 more tools
ask-gemini read-only
Gemini CLI with @ file syntax; 1M+ token context; live progressive output via stream-json
ask-gemini-edit writes
Structured OLD/NEW code edit blocks from Gemini
fetch-chunk read-only
Retrieve chunks from cached large responses
get-usage-stats read-only
Per-session token totals, fallback counts, and breakdowns by provider/model; in-memory only
diagnose read-only
Self-diagnosis: Node version, PATH resolution, provider CLI presence and versions; read-only
ping read-only
Connection test

Use cases

Have Codex or another model review a diff or the last commit for issues the primary model missed
Send an architecture or plan to another model for critique, alternatives, and trade-off analysis
Get an independent second opinion on an approach before committing to it
Read an entire codebase in one call with Gemini or Antigravity (1M+ token context)
Route reviews through Ollama so nothing leaves the machine
Compare several models side by side on one prompt with multi-llm

Supported clients

Claude Code
Claude Desktop
Cursor
Codex
PiPartial support

Listed from the project's documentation, not tested by this site.

Overview

Ask LLM is a unified MCP server that lets Claude Code, Codex CLI, Cursor, Claude Desktop, and other MCP clients consult Gemini, Codex, Claude, Grok, Ollama, and Antigravity over standard MCP for code review, plan debate, and second opinions. It supports multi-turn sessions with a structured AskResponse, plus diagnostics and usage stats, but requires you to install and authenticate the provider CLIs or supply the relevant keys. It is a third-party, unofficial project.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.4k · Tools 2 Compare with this →
Compare with other MCP servers

More servers from this repository

Ask Antigravity MCP ServerAsk LLM Claude MCP ServerAsk Codex MCP ServerAsk GeminiAsk Grok MCP ServerAsk Ollama MCP Server

Data synced 2026-09-25 Read the FMRS scoring method