MCP Tools

SkillAI & models

MCP server management — discover, connect, and use Model Context Protocol tools

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the MCP Tools skill

What this skill tells your AI

The instructions your AI receives, as published by neuron-mr-white/unipi in packages/mcp/skills/mcp/SKILL.md and read by ahel’s review.

MCP (Model Context Protocol) servers provide external tools that extend pi's capabilities. Tools from MCP servers are named {serverName}__{toolName} — e.g., github__search_code.

Adding Servers

Use /unipi:mcp-add to browse the catalog of 7,800+ MCP servers and add them interactively. The split-pane overlay lets you:

  • Browse: Search the cached server catalog by name, description, or category
  • Select: Pick a server to get a pre-filled config template
  • Custom: Press c for an empty editor to add unlisted servers manually
  • Save: Press Enter or Ctrl+S to validate and save

Managing Servers

Use /unipi:mcp-settings to manage configured servers:

  • Toggle: Press Space to enable/disable a server
  • Delete: Press d then y to remove a server
  • Scope: Press g for global config, p for project config
  • Sync: Press s to refresh the catalog from GitHub

Quick Status

Use /unipi:mcp-status for a text summary of all servers and their status.

Config Hierarchy

MCP config supports global defaults with project overrides:

~/.unipi/config/mcp/              ← Global defaults (shared across all projects)
{project}/.unipi/config/mcp/      ← Project overrides (when present)

Merge rules:

  1. Server only in global → used normally
  2. Server only in project → used normally
  3. Server in both → project wins entirely
  4. enabled: false in project metadata → disabled even if defined globally

Config Files

Each level has three files:

  • mcp-config.json — Server definitions (standard MCP format, portable)
  • config.json — Metadata (enabled/disabled, sync prefs)
  • auth.json — Sensitive env vars (optional, chmod 600)

Example mcp-config.json

{
  "mcpServers": {
    "github": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "GITHUB_PERSONAL_ACCESS_TOKEN", "ghcr.io/github/github-mcp-server"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_xxxx" }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/pi/projects"]
    }
  }
}

Using MCP Tools

Once an MCP server is running, its tools are available as pi tools. They're named with the pattern {serverName}__{toolName}.

For example, if you add the GitHub MCP server, you'll have tools like:

  • github__search_code
  • github__create_issue
  • github__list_pull_requests

You can call these tools directly in conversations.

Troubleshooting

Server won't start

  • Check /unipi:mcp-status for error details
  • Verify the command exists: which npx or which docker
  • Check that required env vars are set in the config

Tools not appearing

  • Ensure the server status shows "running"
  • Check if the server supports the MCP tools protocol
  • Try restarting pi after adding a new server

Config issues

  • Validate JSON syntax in your config files
  • Check file permissions (mcp-config.json should be readable)
  • Use /unipi:mcp-settings to view current configuration

Catalog sync issues

  • Run /unipi:mcp-sync to force a refresh
  • Check network connectivity to GitHub
  • The seed catalog (49 servers) is available offline as fallback

Signals

GitHub stars
64
Forks
15
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
mcp-neuron-mr-white
Source
github.com/neuron-mr-white/unipi