MCP Directory

OpenClaw MCP Server

Secure bridge between Claude.ai/Desktop and a self-hosted OpenClaw AI assistant, with OAuth2 and multi-instance routing.

Unverified
stdio (local)
OAuth
TypeScript
View repo 173 Website npm

Add to your client

Copy the config for your MCP client and paste it into its config file.

Install / run
npx openclaw-mcp

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "openclaw-mcp-server": {
      "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"
      }
    }
  }
}

Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf

Before you start

  • Node.js >= 20
  • An OpenClaw gateway running with the HTTP API enabled (gateway.http.endpoints.chatCompletions.enabled = true)
  • OPENCLAW_GATEWAY_TOKEN for the OpenClaw gateway
  • For remote (Claude.ai) use: AUTH_ENABLED with MCP_CLIENT_ID / MCP_CLIENT_SECRET, MCP_ISSUER_URL and TRUST_PROXY=1 behind a reverse proxy
  • Docker (optional, for the GHCR image deployment)

About OpenClaw MCP Server

An MCP bridge server that connects Claude.ai or Claude Desktop to a self-hosted OpenClaw AI assistant gateway. It enables Claude to delegate chat and long-running tasks to one or more OpenClaw instances, with OAuth 2.1 auth for remote use and isolated per-instance configuration for multi-gateway orchestration.

Tools & capabilities (7)

openclaw_chat

Send messages to OpenClaw and get responses (synchronous). Accepts an optional instance parameter to target a specific gateway.

openclaw_status

Check OpenClaw gateway health. Accepts an optional instance parameter.

openclaw_instances

List all configured OpenClaw instances (tokens are never exposed).

openclaw_chat_async

Queue a message for a long-running operation and get a task_id immediately. Accepts an optional instance parameter.

openclaw_task_status

Check task progress and get results.

openclaw_task_list

List all tasks with filtering.

openclaw_task_cancel

Cancel a pending task.

When to use it

  • Delegating tasks from the Claude.ai web UI to a self-hosted OpenClaw bot that can spin up Claude Code to fix issues
  • Orchestrating one AI assistant (Claude) to drive another (OpenClaw)
  • Running long async jobs against OpenClaw and polling for results via task tools
  • Routing requests across multiple OpenClaw gateways (prod, staging, dev) from a single MCP bridge
  • Securely exposing a self-hosted OpenClaw assistant to Claude.ai over OAuth 2.1 + HTTPS

Security notes

Always enable authentication in production: set AUTH_ENABLED=true with MCP_CLIENT_ID and a strong MCP_CLIENT_SECRET (e.g. openssl rand -hex 32). Restrict origins via CORS_ORIGINS (e.g. https://claude.ai). When running behind a reverse proxy you must set MCP_ISSUER_URL to your public HTTPS URL and TRUST_PROXY=1, otherwise OAuth metadata and the /token rate limiter break. The remote connector URL must end with /mcp (the Streamable HTTP endpoint). OAuth 2.1 is used for the remote (Claude.ai) transport; the local stdio transport authenticates to the OpenClaw gateway via OPENCLAW_GATEWAY_TOKEN. Tokens are never exposed via openclaw_instances. The Docker image runs read_only with no-new-privileges.

OpenClaw MCP Server FAQ

Do I need Docker?

No. Docker is the recommended path for remote (Claude.ai) deployments via the GHCR image, but you can also run it locally with `npx openclaw-mcp` for Claude Desktop, or remotely with `npx openclaw-mcp --transport http --port 3000`.

Why does my Claude.ai connector return 404 after OAuth?

The connector URL must end with /mcp, which is the Streamable HTTP endpoint. A bare domain hits the server root and returns 404 after OAuth completes.

What must I set behind a reverse proxy?

Set MCP_ISSUER_URL (or --issuer-url) to your public HTTPS URL so OAuth metadata is correct, and TRUST_PROXY=1 (or --trust-proxy 1) so express-rate-limit accepts the proxy's X-Forwarded-For header instead of crashing /token.

Is SSE still supported?

Yes, but it is deprecated since v1.5.0. The primary transport is now Streamable HTTP (--transport http). --transport sse still works as a deprecated alias and both transports are served simultaneously.

How does multi-instance mode work?

Set OPENCLAW_INSTANCES to a JSON array of named gateways (each with url, token, and optional default). Every tool accepts an optional `instance` parameter; when omitted, the default instance is used. Existing single-instance deployments work unchanged.

#openclaw#ai-assistant#oauth#self-hosted#docker#orchestration#typescript#bridge

Alternatives to OpenClaw MCP Server

Compare all alternatives →
Memory (Knowledge Graph)
AI, Data & Knowledge
74k

Official MCP server providing persistent, file-backed knowledge-graph memory across sessions.

Verified
stdio (local)
No auth
TypeScript
9 tools
Updated 5 months agoRepo
Sequential Thinking
AI, Data & Knowledge
62k

Structured step-by-step reasoning tool for breaking problems into revisable thought sequences.

Verified
stdio (local)
No auth
TypeScript
1 tool
Updated 6 months agoRepo
AWS Knowledge MCP
AI, Data & Knowledge
7.0k

Fully managed remote server for AWS docs, blogs, What's-New and Well-Architected guidance — no key.

Verified
HTTP (remote)
No auth
Hosted
6 tools
Updated 5 months agoRepo

Compare OpenClaw MCP Server with:

vs Memory (Knowledge Graph)vs Sequential Thinkingvs AWS Knowledge MCPvs MCP Server Chart