OpenClaw MCP Server

Community

MCP server securely bridging Claude with a self-hosted OpenClaw assistant.

How to install ↓ Compare GitHub source repository ↗

Category: Other
Stars: ★ 186 Very popular
Transport: stdio (local process) · Streamable HTTP
Runtime: Node.js 20+ · Docker
Credentials: API key / credential required
License: MIT
Last commit: 23 days ago
Tools: 7

63FMRS · C

The server provides seamless integration with OpenClaw, supporting both sync and async tasks and multi-instance. Security is emphasized with OAuth and CORS controls. Deployment requires careful configuration.

Strongest · Documentation 17/20 Weakest · Reliability 6/20

Reliability

6/20

Security and permissions

13/20

Maintenance

15/20

Documentation

17/20

Setup experience

12/20

Why each score

Reliability 6/20: Evidence: the README lists the sync/async tool set and both transport modes, and the manifest confirms the npm 1.7.0 package and OCI image exist. Deductions: no source code, tests, or CI logs were supplied, so the MCP init, tool-list handshake, and alignment of declared tools with real behavior cannot be verified; the manifest declares the OCI package as stdio transport pinned at 1.5.0, contradicting the README's HTTP-first bridge deployment at 1.7.0. Uncertainty only lowers the score, hence 6.
Security and permissions 13/20: Evidence: the manifest flags OPENCLAW_GATEWAY_TOKEN as a secret; the README mandates OAuth 2.1 in production, defaults CORS to off since 1.7.0, scopes async tasks to the originating connection, states openclaw_instances never exposes tokens, runs containers read_only with no-new-privileges, and publishes a threat model; install examples use placeholders only, with no real secrets. Deductions: the implementation cannot be audited from the supplied material, and the bridge's remote-command surface is only as safe as the gateway token (disclosed but unverifiable), so full marks are withheld; score 13.
Maintenance 15/20: Evidence: MIT license, the version line advanced to 1.7.0 with upgrade notes, CI badge, GHCR releases, SECURITY.md, and layered documentation; the manifest independently confirms published npm and OCI packages. Deductions: commit cadence, issue-response timeliness, and dependency-update history cannot be verified from the supplied material, so a conservative 15 is given.
Documentation 17/20: Evidence: the README covers Docker/local/remote setups, multi-instance mode, SSE→HTTP migration, architecture diagrams, tool tables, security and upgrade sections, links layered docs (installation, configuration, deployment, threat model, logging, development, SECURITY), and includes troubleshooting for ERR_ERL_UNEXPECTED_X_FORWARDED_FOR and Unregistered redirect_uri. Deductions: no explicit limits/cost guidance and the full OAuth flow details are deferred to external docs, so 17.
Setup experience 12/20: Evidence: docker-compose example, Claude Desktop config JSON, Claude.ai connector steps, and environment variable tables are provided. Deductions: the flow is multi-step and fragile — the connector URL must end with /mcp, redirect_uri matching is exact, and MCP_ISSUER_URL plus TRUST_PROXY are required behind a reverse proxy; the manifest's OCI entry (1.5.0/stdio) conflicts with the README's 1.7.0/HTTP story and can confuse installs; no execution evidence, static cap 15, so score 12.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

Users who want to integrate self-hosted OpenClaw with Claude.
Users managing multiple OpenClaw instances from a unified interface.

Not for

Users without an OpenClaw instance.
Users requiring unauthenticated production deployments.

Required permissions

Access to the OpenClaw gateway HTTP API.
Requires a gateway token (Bearer) for authentication.

Risks and side effects

Security risks in production if not properly secured (authentication must be enabled).
OAuth callback exposure if misconfigured; precise redirect URIs are recommended.
Exposing gateway URL and secrets could lead to security issues.

Setup

Before you start

Runtime：Node.js 20+ · Docker

OPENCLAW_URL required URL of the OpenClaw gateway (OpenAI-compatible HTTP API), default http://127.0.0.1:18789; required.

OPENCLAW_GATEWAY_TOKEN requiredsecret Bearer token for OpenClaw gateway authentication; obtain it from your self-hosted OpenClaw gateway; required secret.

MCP_CLIENT_SECRET optionalsecret OAuth client secret, generate with openssl rand -hex 32; secret.

Other optional settings (9)

OPENCLAW_MODEL optional Model name for chat completions, e.g. 'openclaw' or 'openclaw/<agentId>'; default 'openclaw'; optional.

OPENCLAW_TIMEOUT_MS optional Request timeout in milliseconds; default 120000; optional.

OPENCLAW_INSTANCES optional JSON array of instances (name, url, token, etc.) for multi-instance mode; optional.

AUTH_ENABLED optional Set to true to enable OAuth 2.1 auth in HTTP mode; recommended for production; optional.

MCP_CLIENT_ID optional OAuth client ID used by the Claude.ai connector (e.g. 'openclaw'); used in HTTP mode.

MCP_ISSUER_URL optional Public HTTPS URL required behind a reverse proxy so OAuth metadata doesn't advertise localhost.

MCP_REDIRECT_URIS optional OAuth redirect URIs; for Claude.ai use https://claude.ai/api/mcp/auth_callback,https://claude.com/api/mcp/auth_callback.

TRUST_PROXY optional Set to 1 behind a reverse proxy so express-rate-limit accepts the X-Forwarded-For header.

CORS_ORIGINS optional Allowed CORS origins (e.g. https://claude.ai); off by default; set only when a browser client connects directly.

For Claude Desktop (local): run npx openclaw-mcp, then add the binary configuration to claude_desktop_config.json with environment variables. For Claude.ai (remote): use Docker or CLI with HTTP transport, enable authentication, and configure OAuth redirect URIs.

claude_desktop_config.json

{
  "mcpServers": {
    "openclaw": {
      "command": "npx",
      "args": [
        "openclaw-mcp"
      ],
      "env": {
        "OPENCLAW_URL": "http://127.0.0.1:18789",
        "OPENCLAW_GATEWAY_TOKEN": "your-gateway-token",
        "OPENCLAW_MODEL": "openclaw",
        "OPENCLAW_TIMEOUT_MS": "300000"
      }
    }
  }
}

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

.vscode/mcp.json

{
  "servers": {
    "openclaw": {
      "command": "npx",
      "args": [
        "openclaw-mcp"
      ],
      "env": {
        "OPENCLAW_URL": "http://127.0.0.1:18789",
        "OPENCLAW_GATEWAY_TOKEN": "your-gateway-token",
        "OPENCLAW_MODEL": "openclaw",
        "OPENCLAW_TIMEOUT_MS": "300000"
      }
    }
  }
}

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

Terminal

claude mcp add openclaw -e OPENCLAW_URL=http://127.0.0.1:18789 -e OPENCLAW_GATEWAY_TOKEN=your-gateway-token -e OPENCLAW_MODEL=openclaw -e OPENCLAW_TIMEOUT_MS=300000 -- npx openclaw-mcp

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

Generate and validate client config

Check that it works

The client's tool list should show the seven openclaw_* tools such as openclaw_chat, openclaw_status, and openclaw_instances. Running openclaw_status and getting gateway health info confirms the bridge reaches the OpenClaw gateway.

Troubleshooting

Ensure the connector URL ends with `/mcp`, otherwise a 404 may occur after OAuth.
Ensure `MCP_ISSUER_URL` is set to your public HTTPS URL and `TRUST_PROXY` is enabled to avoid authentication issues.
If you see 'Unregistered redirect_uri', check that `MCP_REDIRECT_URIS` exactly matches Claude's callback path.

Things to try

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

Ask OpenClaw 'Deploy status?' and relay its answer to me.
Check the health of the OpenClaw gateway.
List all configured OpenClaw instances.
Queue the message 'Run tests' to the dev instance and report the task's progress.

Tools 7

openclaw_chat writes

Send messages to OpenClaw and get responses.

openclaw_status read-only

Check OpenClaw gateway health.

openclaw_instances read-only

List all configured OpenClaw instances.

openclaw_chat_async writes

Queue a message and get a task ID immediately.

openclaw_task_status read-only

Check task progress and get results.

openclaw_task_list read-only

List your tasks with filtering.

openclaw_task_cancel writes

Cancel a pending task.

Use cases

Delegating tasks from Claude to OpenClaw assistants.

Orchestrating multiple OpenClaw instances (e.g., prod, staging, dev) from a single MCP server.

Running long-running operations asynchronously without blocking responses.

Supported clients

Claude Desktop

Claude.ai

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

Overview

The OpenClaw MCP server is a secure bridge connecting Claude.ai or Claude Desktop to a self-hosted OpenClaw assistant via OAuth 2.1 authentication. It allows Claude to delegate tasks to OpenClaw instances, supporting both synchronous and asynchronous operations. The server supports multi-instance mode for orchestrating multiple OpenClaw gateways from a single MCP server.

Similar servers

DuckDuckGo Search MCP Server 74 · B

MCP server providing web search and content fetching via DuckDuckGo

★ 1.5k · Tools 2 Compare with this →

Basic Memory 72 · B

Your AI never forgets again.

★ 4.1k · Tools 21 Compare with this →

Telegram MCP Server 70 · B

Telegram MCP server powered by Telethon to let MCP clients read chats, manage groups, and send/modify messages, media, contacts, and settings.

★ 1.8k · Tools 76 Compare with this →

Hostinger API MCP Server 70 · B

Official MCP server for Hostinger API.

★ 160 · Tools 100 Compare with this →

Compare with other MCP servers

Source revision 3514ac1aa8a1 Data synced 2026-10-12 Read the FMRS scoring method