OpenClaw MCP Server for Claude
Connect Claude, Codex, Claude Code, and other MCP clients to OpenClaw Gateway-backed channel conversations with the official `openclaw mcp serve` bridge, plus manage OpenClaw-saved outbound MCP server definitions.
Open the source and read safety notes before installing.
Safety notes
- `messages_send` can send text back through an existing OpenClaw channel route, so confirm the target conversation before approving replies.
- `permissions_respond` can resolve exec or plugin approval requests observed while the bridge is connected.
- The live event queue and Claude channel notifications exist only while the MCP bridge session is connected; use transcript reads for older history.
- Remote Gateway URLs, token files, and password files should be treated as secrets and scoped to trusted machines.
- OpenClaw channel safety remains controlled by the underlying Gateway, pairing policy, sender allowlists, and sandbox settings.
Privacy notes
- Routed conversation titles, transcript history, non-text message metadata, sender identifiers, account IDs, thread IDs, and live inbound events may be exposed to the MCP client and model.
- Approval request details can reveal command text, plugin names, paths, or operational intent while the bridge is connected.
- MCP client logs, transcript exports, and model prompts may retain OpenClaw conversation content outside the Gateway.
- Do not paste Gateway tokens, password file contents, channel credentials, or MCP server secrets into prompts, public issues, screenshots, or committed config.
Prerequisites
- OpenClaw CLI installed from npm and a working OpenClaw Gateway.
- An MCP client such as Claude Code, Codex, Claude Desktop, or another stdio-capable host.
- Existing OpenClaw session route metadata for conversations you want exposed through MCP.
- Gateway token or password files when connecting to a remote OpenClaw Gateway.
- Channel pairing, sender allowlists, and trust policies reviewed before exposing live messaging routes.
Schema details
- Install type
- cli
- Troubleshooting
- No
- Scope
- Source repo
- Estimated setup
- 15 minutes
- Difficulty
- intermediate
- Website
- https://openclaw.ai
Full copyable content
openclaw mcp serveAbout this resource
Overview
OpenClaw ships an official MCP surface through the openclaw mcp CLI. The most
important path for Claude and other external clients is openclaw mcp serve,
which starts a stdio MCP server, connects to an OpenClaw Gateway over WebSocket,
and exposes routed OpenClaw channel conversations as MCP tools.
Use this when an MCP client should inspect and reply to OpenClaw-backed
conversations without running a separate bridge for each channel. OpenClaw also
uses the same openclaw mcp namespace for its client-side MCP registry, where
operators can save, inspect, probe, and update third-party MCP server
definitions for OpenClaw-managed runtimes.
Features
- Start OpenClaw as a stdio MCP server with
openclaw mcp serve. - Connect the bridge to a local or remote OpenClaw Gateway.
- List routed conversations with
conversations_list. - Fetch a single conversation with
conversation_get. - Read recent transcript history with
messages_read. - Inspect non-text transcript message metadata with
attachments_fetch. - Poll or wait for live inbound events with
events_pollandevents_wait. - Send a text reply through an existing route with
messages_send. - List and resolve live exec or plugin approvals with
permissions_list_openandpermissions_respond. - Enable Claude-specific channel notifications for live messages and permission events.
- Manage OpenClaw-saved outbound MCP server definitions with
openclaw mcp add,set,configure,tools,login,logout,status,doctor,probe,reload, andunset. - Inspect the MCP registry from the OpenClaw Control UI at
/mcp.
Installation
Install the OpenClaw CLI:
npm install -g openclaw@latest
Configure a local MCP client to launch the OpenClaw bridge:
{
"mcpServers": {
"openclaw": {
"command": "openclaw",
"args": ["mcp", "serve"]
}
}
}
For a remote Gateway, pass a WebSocket URL and a token or password file:
openclaw mcp serve --url wss://gateway-host:18789 --token-file ~/.openclaw/gateway.token
openclaw mcp serve --url wss://gateway-host:18789 --password-file ~/.openclaw/gateway.password
Use openclaw mcp serve --claude-channel-mode off when a generic MCP client
should use only standard MCP polling tools instead of Claude-specific channel
notifications.
Bridge Tools
| Tool | Purpose |
|---|---|
conversations_list |
List recent session-backed conversations with route metadata |
conversation_get |
Read one routed conversation by session key |
messages_read |
Read recent transcript messages for a conversation |
attachments_fetch |
Extract non-text transcript message metadata |
events_poll |
Read live queued events since a numeric cursor |
events_wait |
Long-poll for the next matching live event |
messages_send |
Send a text reply through the existing conversation route |
permissions_list_open |
List live approval requests seen by the bridge |
permissions_respond |
Allow or deny one observed approval request |
MCP Registry Commands
The non-serve commands make OpenClaw act as a client-side MCP registry. They
save and inspect server definitions under mcp.servers in OpenClaw config.
They do not expose OpenClaw over MCP by themselves.
Common registry commands:
openclaw mcp list
openclaw mcp status --verbose
openclaw mcp doctor --probe
openclaw mcp probe context7 --json
openclaw mcp add memory --command npx --arg -y --arg @modelcontextprotocol/server-memory
openclaw mcp set docs '{"url":"https://mcp.example.com","transport":"streamable-http"}'
openclaw mcp configure docs --auth oauth --oauth-scope 'docs.read'
openclaw mcp login docs
openclaw mcp unset context7
Use probe or doctor --probe when you need live proof that a saved MCP server
connects and lists capabilities.
Use Cases
- Let Claude Code or Codex inspect OpenClaw conversations that already have Gateway route metadata.
- Read recent transcript history before responding to a channel thread.
- Watch live inbound events from OpenClaw-backed channels while an MCP client is connected.
- Send a text reply through the same recorded route after human review.
- Surface live OpenClaw approval requests inside an MCP client session.
- Maintain a central registry of MCP servers that OpenClaw-managed runtimes can later consume.
Safety and Privacy
OpenClaw MCP inherits the authority of the connected Gateway and its configured channels. Sender allowlists, pairing policies, sandboxing, and channel trust boundaries still belong in OpenClaw Gateway configuration. The bridge only sends text through existing routes, but those routes may point at real messaging surfaces.
The live event queue starts when the bridge connects and disappears when the MCP
client disconnects. Use messages_read for durable transcript history. Approval
state is also live-session scoped, so permissions_list_open only includes
requests observed while the bridge was connected.
For remote Gateways, keep token and password files out of prompts, screenshots, logs, and commits. Before exposing OpenClaw beyond localhost, review the official Gateway security and exposure runbooks.
Source Review
Verified on 2026-06-18:
- OpenClaw's MCP documentation describes
openclaw mcp serveas the path for running OpenClaw as an MCP server and the otheropenclaw mcpsubcommands as the client-side registry path. - The documentation lists the bridge tools for conversations, transcript reads, live event polling, message sends, and approval handling.
- The documentation describes Claude-specific channel notifications and notes that generic MCP clients should rely on the standard polling tools.
- The OpenClaw npm package is published as
openclawwith the upstreamopenclaw/openclawrepository.
Duplicate Check
No dedicated OpenClaw MCP server entry was found in content/mcp. Existing
OpenClaw entries in content/skills are skill packs, not the official MCP
bridge or MCP registry documentation.
Source citations
Add this badge to your README
How it compares
OpenClaw MCP Server for Claude side by side with its closest alternative on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.
| Field | OpenClaw MCP Server for Claude Connect Claude, Codex, Claude Code, and other MCP clients to OpenClaw Gateway-backed channel conversations with the official `openclaw mcp serve` bridge, plus manage OpenClaw-saved outbound MCP server definitions. Open dossier | Archestra MCP Platform AGPL-licensed MCP-native platform with a private MCP registry, MCP gateway, Kubernetes MCP orchestrator, access control, credential resolution, observability, and deterministic tool guardrails for shared AI deployments. Open dossier |
|---|---|---|
| Trust | ||
| Install risk | Review first | Review first |
| Notes | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ |
| Category | mcp | mcp |
| Source | source-backed | source-backed |
| Author | OpenClaw | archestra-ai |
| Added | 2026-06-18 | 2026-06-06 |
| Platforms | Claude CodeCodexCursorClaude Desktop | Claude CodeClaude Desktop |
| Source repo | — | — |
| Safety notes | ✓`messages_send` can send text back through an existing OpenClaw channel route, so confirm the target conversation before approving replies. `permissions_respond` can resolve exec or plugin approval requests observed while the bridge is connected. The live event queue and Claude channel notifications exist only while the MCP bridge session is connected; use transcript reads for older history. Remote Gateway URLs, token files, and password files should be treated as secrets and scoped to trusted machines. OpenClaw channel safety remains controlled by the underlying Gateway, pairing policy, sender allowlists, and sandbox settings. | ✓Archestra is a platform/control-plane entry, not a single-purpose local MCP helper; admins can expose many MCP servers, agents, tools, and credentials through one gateway. The upstream quickstart mounts the host Docker socket so the platform can run MCP servers; treat that as highly privileged host access and avoid using it on sensitive machines without isolation. Self-hosted MCP servers may run as Kubernetes workloads with injected environment variables, secrets, images, network policies, and restart controls. Tool assignments, gateway visibility, credential resolution, custom headers, and load-tools-on-demand settings should be reviewed per team and environment. Deterministic tool guardrails can reduce some unsafe tool chains, but they depend on correct policies and do not make untrusted MCP servers safe by default. |
| Privacy notes | ✓Routed conversation titles, transcript history, non-text message metadata, sender identifiers, account IDs, thread IDs, and live inbound events may be exposed to the MCP client and model. Approval request details can reveal command text, plugin names, paths, or operational intent while the bridge is connected. MCP client logs, transcript exports, and model prompts may retain OpenClaw conversation content outside the Gateway. Do not paste Gateway tokens, password file contents, channel credentials, or MCP server secrets into prompts, public issues, screenshots, or committed config. | ✓MCP server definitions, tool schemas, gateway tokens, upstream credentials, OAuth tokens, API keys, custom headers, logs, traces, and tool results may be stored or processed by the platform. Built-in observability, LLM proxy, chat, agents, and policy features can reveal prompts, tool arguments, tool outputs, token usage, user identities, team membership, and trace metadata. Registry entries and installations can use personal, team-scoped, or shared credentials; choose the narrowest scope that matches the use case. When external MCP clients call an Archestra gateway, downstream tool results can still be sent by the MCP client to the configured model provider. |
| Prerequisites |
|
|
| Install | | |
| Config | | |
| Citations | ||
| Claim | Unclaimed | Unclaimed |
Signals
Loading live community signals…
A short, calm digest of reviewed Claude resources. Unsubscribe any time.