FrootAI MCP for Python

Give MCP-compatible agents a local Python process with 62 FrootAI tools, 4 resources, 6 prompts, and trust-gated federation.

Python product page · Setup guide · PyPI · MCP guide

Choose Python MCP or the SDK

Python MCP runs locally over stdio. Installing it does not start a daemon, publish an endpoint, deploy cloud resources, or grant a client blanket permission to modify files.

Five steps to first value

1. Install and verify

python -m pip install --upgrade frootai-mcp
frootai-mcp-py

Requirements: Python 3.10 or newer. Press Ctrl+C after confirming that the server starts; an MCP client normally owns the process lifecycle.

2. Connect an MCP client

VS Code and GitHub Copilot — .vscode/mcp.json:

{
  "servers": {
    "frootai-python": {
      "type": "stdio",
      "command": "frootai-mcp-py"
    }
  }
}

Claude Desktop or Cursor:

{
  "mcpServers": {
    "frootai-python": {
      "command": "frootai-mcp-py"
    }
  }
}

If the executable is not on PATH, use the absolute path from the same Python environment in which the package was installed.

3. Ask for grounded discovery

Search FrootAI knowledge for secure RAG patterns.
Find the best two Solution Plays, compare them,
and show the relevant WAF security guidance.

A typical sequence is:

search_knowledge → semantic_search_plays → compare_plays
                 → get_play_detail → get_waf_guidance

4. Wire, scaffold, and evaluate

wire_play → validate_manifest → preview_scaffold
          → evaluate_quality → agent_review → agent_tune

Use preview and validation before any write-oriented operation. wire_play, scaffold_play, smart_scaffold, and live evaluation expose their operating boundary through MCP annotations and return values.

5. Use the server from Python when needed

from frootai_mcp.server import mcp

mcp.run(transport="stdio")

Tools are async Python functions and can also be called directly:

import asyncio
from frootai_mcp.server import search_knowledge, wire_play

async def main() -> None:
    results = await search_knowledge(query="RAG architecture")
    manifest = await wire_play("01")
    print(results)
    print(manifest)

asyncio.run(main())

For application-level APIs rather than an MCP server process, prefer the frootai SDK.

Capability map — 62 tools

MCP resources

MCP prompts

What ships inside

The Python and Node MCP packages currently register the same 62 primary tool names. Transport implementation, packaging, and runtime dependencies differ; treat tools/list from the running server as authoritative rather than assuming byte-identical behavior.

search_knowledge → semantic_search_plays → get_play_detail
agent_build → agent_review → agent_tune
wire_play → validate_manifest → evaluate_quality
preview_scaffold → scaffold_play → scaffold_status
fai_discover_mcp → fai_trust_query → fai_attach_mcp
                 → fai_invoke_via → fai_detach_mcp

Safety and operating boundaries

Verify and develop

cd python-mcp
python -m pip install --upgrade build pytest
python -m pytest tests -v
python -m build

Live provider suites are opt-in and require their documented environment variables.

Related packages

Links

• Python product page
• Python setup
• FrootAI MCP guide
• FAI Orchard
• Security and trust
• PyPI package
• Issues

License

MIT © 2026 FrootAI.