Skip to content

MCP Installation

Add RFlect's MCP server to your MCP client. The server entry point is rflect-mcp/server.py. RFlect runs identically on Linux, macOS, and Windows — the only per-OS differences are the Python path and where your client keeps its config.

1. Install RFlect + MCP dependencies

git clone https://github.com/RFingAdam/RFlect.git
cd RFlect
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
pip install -r rflect-mcp/requirements.txt

Your Python interpreter is then …/RFlect/.venv/bin/python.

git clone https://github.com/RFingAdam/RFlect.git
cd RFlect
py -m venv .venv
.venv\Scripts\Activate.ps1
pip install -e .
pip install -r rflect-mcp\requirements.txt

Your Python interpreter is then …\RFlect\.venv\Scripts\python.exe.

No API key required

RFlect makes no LLM/API calls and needs no key or subscription. The MCP server runs fully offline.

2. Register with your MCP client

Claude Code

Config file location:

OS Path
Linux / macOS ~/.claude/settings.json
Windows %USERPROFILE%\.claude\settings.json

(Or use a project-local .mcp.json in your repo root, which works the same on every OS.)

{
  "mcpServers": {
    "rflect": {
      "command": "/absolute/path/to/RFlect/.venv/bin/python",
      "args": ["/absolute/path/to/RFlect/rflect-mcp/server.py"]
    }
  }
}

Use forward slashes (or escaped \\) in JSON, and the Scripts venv path:

{
  "mcpServers": {
    "rflect": {
      "command": "C:/Users/you/RFlect/.venv/Scripts/python.exe",
      "args": ["C:/Users/you/RFlect/rflect-mcp/server.py"]
    }
  }
}

Restart Claude Code; you should see 41 RFlect tools available.

Cline (VS Code)

.cline/mcp_settings.json in your project root (same shape on every OS — just swap the Python path):

{
  "mcpServers": {
    "rflect": {
      "command": "/absolute/path/to/RFlect/.venv/bin/python",
      "args": ["/absolute/path/to/RFlect/rflect-mcp/server.py"],
      "env": {}
    }
  }
}

{
  "mcpServers": {
    "rflect": {
      "command": "C:/Users/you/RFlect/.venv/Scripts/python.exe",
      "args": ["C:/Users/you/RFlect/rflect-mcp/server.py"],
      "env": {}
    }
  }
}

Continue (VS Code)

Continue uses ~/.continue/config.json (Linux/macOS) or %USERPROFILE%\.continue\config.json (Windows):

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "/absolute/path/to/RFlect/.venv/bin/python",
          "args": ["/absolute/path/to/RFlect/rflect-mcp/server.py"]
        }
      }
    ]
  }
}

Other MCP clients

Any client supporting stdio MCP can host the server. The launch command is always your venv Python plus the server path — no special args:

<venv-python> /path/to/rflect-mcp/server.py

3. Verify

Ask the assistant "What RFlect tools do you have?" — it should list 41. Or run the smoke test from a shell (works on every OS):

cd RFlect
.venv/bin/python -c "import sys, os; sys.path.insert(0, os.path.abspath('rflect-mcp')); import server; print('tools:', len(server.mcp._tool_manager._tools))"

cd RFlect
.venv\Scripts\python.exe -c "import sys, os; sys.path.insert(0, os.path.abspath('rflect-mcp')); import server; print('tools:', len(server.mcp._tool_manager._tools))"

Expected: tools: 41.

Troubleshooting

If the server fails to connect, see Troubleshooting. The most common cause is pointing command at a Python that doesn't have the rflect-mcp/requirements.txt dependencies installed — use the venv interpreter, not the system python.