io.github.fernandoludvig/rdstation-crm-mcp

rdstation-crm-mcp

An open-source MCP server for RD Station CRM — the leading CRM in Brazil and Latin America. Manage contacts, deals, tasks and notes, and get pipeline health reports, straight from Claude or any MCP-compatible client.

"How's my sales pipeline this month?" → stage-by-stage totals, win rate, and the deals going stale.

Why

RD Station CRM is huge in the LatAm market, but had no open-source MCP server. This project connects it to the MCP ecosystem so AI agents can work your pipeline: qualifying leads, moving deals, scheduling follow-ups, and answering questions about your sales data in natural language.

Quick start

1. Get your instance token in RD Station CRM: Profile → Products and integrations → Instance token.
2. Add the server to your MCP client.

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "rdstation-crm": {
      "command": "npx",
      "args": ["-y", "rdstation-crm-mcp"],
      "env": {
        "RDSTATION_CRM_TOKEN": "your-instance-token"
      }
    }
  }
}

Claude Code:

claude mcp add rdstation-crm -e RDSTATION_CRM_TOKEN=your-instance-token -- npx -y rdstation-crm-mcp

That's it. Ask Claude something like "list my open deals" or "give me a pipeline overview".

Tools

Design notes

These tools are designed for LLMs, not as a 1:1 API wrapper:

• Names instead of IDs. Stages, pipelines, users and lost reasons can be passed by name; the server resolves them against the account and lists the valid options when something doesn't match.
• Compact responses. List tools return one line per record with the fields that matter, plus explicit pagination hints. Large responses are truncated with guidance instead of flooding the context window.
• Actionable errors. A 401 tells you which env var to check; an unknown stage lists every stage in every pipeline.
• Aggregation where it counts. rdcrm_pipeline_overview answers the questions humans actually ask ("where are deals stuck?") with a single tool call.

Remote deployments (Streamable HTTP)

The default (rdstation-crm-mcp) runs over stdio for a single local user, with the token read once from RDSTATION_CRM_TOKEN. For a team-hosted or registry-listed deployment (e.g. Smithery), run the Streamable HTTP variant instead:

npx -y rdstation-crm-mcp-http

This starts an HTTP server (http://127.0.0.1:8080/mcp by default) implementing the MCP Streamable HTTP transport, with proper session lifecycle (initialize → Mcp-Session-Id → DELETE to close).

Because a hosted server can serve more than one user, there's no single implicit token: each session sends its own Authorization: Bearer <token> header on initialize. A RDSTATION_CRM_TOKEN env var still works as a fallback default for a single-tenant self-hosted setup where every caller shares one CRM account.

Env vars:

curl http://127.0.0.1:8080/health
# {"status":"ok","server":"rdstation-crm-mcp","sessions":0}

Development

git clone https://github.com/fernandoludvig/rdstation-crm-mcp.git
cd rdstation-crm-mcp
npm install
npm test              # unit tests (API mocked with msw)
npm run typecheck
npm run build
RDSTATION_CRM_TOKEN=xxx npx @modelcontextprotocol/inspector node dist/index.js

The HTTP layer (src/client/) is isolated from the tools, with retry and exponential backoff for 429/5xx built in.

Roadmap

• [ ] Organizations and products tools
• [ ] RD Station CRM API v2 support (OAuth) behind the same tool surface
• [x] Streamable HTTP transport for remote deployments
• [ ] Publish to MCP registries (Glama, PulseMCP, Smithery)

Contributions welcome — open an issue first for anything non-trivial.

License

MIT © Fernando Ludvig

Not affiliated with or endorsed by RD Station. Uses the public RD Station CRM API v1.