Skip to main content
Chatblocks is a hosted MCP server. The same endpoint works everywhere:
Use Authorization: Bearer cb_live_... for human-first setup, or agent-created canvases when the agent needs to create a temporary workspace before the human signs in.

Agent discovery files

Agents should start with these machine-readable URLs:
  • agent.md is the compact operating guide.
  • llms.txt is the short discovery index.
  • llms-full.txt is the complete app-domain agent context in one file.
The docs site also exposes Mintlify’s generated AI-readable documentation. Use the app-domain files above for runtime setup because they are maintained with the product routes and include the current agent-first bootstrap flow.

OpenClaw

OpenClaw can connect directly to Chatblocks as a hosted Streamable HTTP MCP server:
For agent-first setup, OpenClaw should call:
Distribution target: ClawHub skill. The submission artifact lives in integrations/agent-platforms/openclaw-chatblocks-skill/.

Poke

Use the Poke CLI:
Poke can also add Chatblocks through the web integrations UI as a remote MCP server. Use https://chatblocks.ai/api/mcp/v1 with the returned apiKey. For agent-first setup, the recipe should call /api/agent/bootstrap, connect with the returned apiKey, and send the claimUrl to the human. Distribution target: Poke Kitchen recipe. The recipe draft lives at integrations/agent-platforms/poke-kitchen-recipe.json.

Other MCP clients

Any remote-HTTP MCP client can use this shape:
If a client only supports stdio, use the Chatblocks CLI proxy and prefer the hosted HTTP transport whenever the client supports it.

Context7

Context7’s quick submission path expects a GitHub repository URL. For a private Chatblocks monorepo, use one of these paths:
  • Publish a public docs-only mirror that contains apps/docs, context7.json, and generated agent docs.
  • Use Context7 private sources through a Pro or Enterprise workspace.
  • Wait until a public SDK, CLI, or integration repo exists, then submit that repo with links back to the hosted docs.
Context7 should be treated as an agent-discovery accelerator, not as the source of truth. The source of truth remains the hosted docs, /agent.md, /llms.txt, /llms-full.txt, and the MCP server’s tools/list.