RCA-MCP Connector

⚠️ Early access. RCA-MCP's backend is live and this connector has been verified end-to-end against it. The api.rca-mcp.com custom domain isn't wired up yet — point RCA_MCP_API_URL at the current backend URL below. Tool schemas and documentation may still change before the first stable public launch.

What is RCA-MCP?

The only MCP server purpose-built for causal Root Cause Analysis. 56 tools covering causal graph construction, 10 RCA model families plus 3 dedicated PyRCA algorithms (Salesforce PyRCA, BSD-3-Clause), multi-model consensus, and PDF/HTML/Excel/Markdown report generation. Works with Claude, Ollama, Groq, OpenAI, Gemini, LangChain, Cursor — 10 providers.

Quick Start (2 minutes)

rca-mcp-connector is a published PyPI package — no clone needed. Point any MCP client at it with uvx (or pip install rca-mcp-connector if you'd rather manage the install yourself):

uvx rca-mcp-connector

Get a free API key at rca-mcp.pages.dev — no credit card required — then set RCA_MCP_API_KEY in your MCP client's config (examples below).

Claude Code Setup

Add to .mcp.json in your workspace root:

{
  "mcpServers": {
    "rca-mcp": {
      "command": "uvx",
      "args": ["rca-mcp-connector"],
      "env": {
        "RCA_MCP_API_URL": "https://rcamcp-production.up.railway.app",
        "RCA_MCP_API_KEY": "your_api_key_here"
      }
    }
  }
}

Ollama Setup

go install github.com/mark3labs/mcphost@latest
mcphost -m ollama:qwen3:14b --config providers/mcp-servers.json

OpenAI Agents SDK

from agents import Agent, MCPServerStdio
import asyncio

async def main():
    async with MCPServerStdio(
        params={
            "command": "uvx",
            "args": ["rca-mcp-connector"],
            "env": {
                "RCA_MCP_API_URL": "https://rcamcp-production.up.railway.app",
                "RCA_MCP_API_KEY": "your_api_key_here",
            },
        }
    ) as rca_server:
        agent = Agent(name="RCA Agent", model="gpt-4o", mcp_servers=[rca_server])
        result = await agent.run("Find the root cause of the API latency spike.")
        print(result.final_output)

asyncio.run(main())

LangChain

from langchain_mcp_adapters.client import MultiServerMCPClient
from langchain_anthropic import ChatAnthropic
from langgraph.prebuilt import create_react_agent
import asyncio

async def main():
    async with MultiServerMCPClient({
        "rca-mcp": {
            "command": "uvx", "args": ["rca-mcp-connector"],
            "env": {
                "RCA_MCP_API_URL": "https://rcamcp-production.up.railway.app",
                "RCA_MCP_API_KEY": "your_api_key_here",
            },
            "transport": "stdio",
        }
    }) as client:
        tools = await client.get_tools()
        agent = create_react_agent(ChatAnthropic(model="claude-sonnet-4-6"), tools)
        result = await agent.ainvoke({"messages": [{"role": "user", "content": "Run an FMEA analysis"}]})
        print(result["messages"][-1].content)

asyncio.run(main())

See providers/ for ready-to-use config templates and full examples (Groq, Gemini, OpenRouter, Claude Desktop).

Third-Party Licences

Algorithms in rca_pyrca_* tools are independently-written adaptations of PyRCA's published methods (Zheng et al. 2023, arXiv:2306.11417), not direct copies of PyRCA source code, per the private API's models/pyrca_adapter.py.

Citing RCA-MCP

@software{rcamcp2026,
  title  = {RCA-MCP: An MCP Server for Causal Root Cause Analysis},
  author = {davetj},
  year   = {2026},
  url    = {https://github.com/dave1362/rca-mcp-connector},
  note   = {v4.1.13}
}