OpenClaw MCP setup - PlatformDTC

PlatformDTC ships a Model Context Protocol server (mcp-platformdtc) that exposes the Agent Gateway as dtc_* tools — the same pattern OpenClaw uses for its other integrations. Register it once per container and the agent can drive the entire platform.

Register the server

Add the server to the OpenClaw container’s config. The container’s scoped sq_agt_* key is passed on the ?key= query string and forwarded to the Gateway as X-Agent-Key:

{
  "mcpServers": {
    "platformdtc": {
      "type": "streamable-http",
      "url": "http://host.docker.internal:18810/mcp?key=sq_agt_xxx",
      "tools": ["dtc_*"]
    }
  }
}

Run the MCP process alongside your gateway (e.g. PM2):

GATEWAY_BASE_URL=https://api.platformdtc.com/api/v1/agent/v1 \
MCP_PLATFORMDTC_PORT=18810 \
python mcp-platformdtc/server.py

How calls flow

The MCP server forwards each dtc_* call 1:1 to its REST route, auto-injects an Idempotency-Key for mutations, and returns the Gateway’s JSON. Because it’s 1:1, anything an agent can do, you can reproduce with curl — handy for debugging.

Golden rules for agents

Async → wait

Any tool that returns a job_id must be followed by dtc_job_wait.

Fresh idempotency key

Pass a new UUID idempotency_key per action.

Respect the approval gate

If a call returns pending_approval, tell the user and wait — don’t retry.

Stay in scope

Call dtc_whoami first; only attempt tools your scopes allow.

Next: the end-to-end Playbook.

​Register the server

​How calls flow

​Golden rules for agents

Register the server

How calls flow

Golden rules for agents