MCP Guides

MCP: Extend Autohand with External Tools

The Model Context Protocol (MCP) lets Autohand connect to external services. Search the web, pull live documentation, run code in sandboxes, convert Figma designs to code, and more. Each MCP server adds new capabilities without changing Autohand itself.

What is MCP?

MCP is an open standard that connects AI agents to external tools and data sources. Think of it like USB for AI: a single protocol that lets any tool talk to any agent. Autohand's MCP client supports three transport types:

  • stdio - Runs a local process and communicates over stdin/stdout. Best for most tools.
  • SSE - Connects to a remote server using Server-Sent Events. Good for shared team servers.
  • HTTP - Sends requests to an HTTP endpoint. Simplest for cloud-hosted services.

When you add an MCP server, its tools become available to the agent automatically. Tools are namespaced as mcp__<server>__<tool> so there are no conflicts between servers.

Quick start

Adding an MCP server takes one command:

# Add a server (stdio transport is the default)
autohand mcp add brave-search npx -y @anthropic/mcp-server-brave-search

# Add a server with HTTP transport
autohand mcp add --transport http context7 https://mcp.context7.com/mcp

# List connected servers
autohand mcp list

# Remove a server
autohand mcp remove brave-search

You can also configure servers in ~/.autohand/config.json for persistent setup:

{
  "mcp": {
    "servers": [
      {
        "name": "brave-search",
        "transport": "stdio",
        "command": "npx",
        "args": ["-y", "@anthropic/mcp-server-brave-search"],
        "env": {
          "BRAVE_API_KEY": "your-api-key"
        }
      }
    ]
  }
}

Guides

Each guide walks through setup, usage patterns, and practical examples for a specific MCP server.

How to search the web

Use Brave Search to look up error messages, find package documentation, and research solutions without leaving the terminal.

Read guide

How to get accurate documentation

Use Context7 to pull current, version-specific docs for any library so the agent codes against real APIs instead of stale training data.

Read guide

How to extract web content

Use Firecrawl to convert any web page into clean markdown. Read full articles, API references, and tutorials directly in your session.

Read guide

How to turn designs into code

Use Figma MCP to read design files and generate pixel-accurate components with correct spacing, colors, and typography.

Read guide

How to run code in a sandbox

Use E2B to execute code in isolated cloud environments. Test scripts, try packages, and process data without touching your local machine.

Read guide

Find more MCP servers

The MCP ecosystem is growing fast. These community-maintained lists collect hundreds of servers for databases, APIs, developer tools, and more.

Curated lists

  • awesome-mcp-servers - The largest community list. Organized by category with popularity rankings.
  • awesome-mcp-servers (appcypher) - Another well-maintained collection with different categorization.
  • Official MCP servers - Reference implementations maintained by the MCP team.
  • mcp.so - Searchable directory with install commands and compatibility info.
  • Glama MCP directory - Curated directory with categories, ratings, and one-click install.
  • Smithery - MCP server registry with hosted versions you can connect to instantly.

Popular categories

Category Examples What they do
Databases PostgreSQL, SQLite, Supabase Query and manage databases directly from the agent
Search Brave Search, Tavily, Exa Search the web for documentation, solutions, references
Documentation Context7, Firecrawl Pull live docs and web content into sessions
Cloud platforms AWS, Cloudflare, Vercel Manage infrastructure and deployments
Design Figma Read design files and generate matching code
Code execution E2B, Docker Run code in sandboxed environments
Monitoring Sentry, Datadog, Grafana Read errors, metrics, and logs for debugging
Version control GitHub, GitLab, Linear Manage issues, PRs, and project boards

Building your own MCP server

If you have an internal API or a custom tool, you can build your own MCP server. The protocol is straightforward and SDKs are available in multiple languages.

# Install the TypeScript SDK
npm install @modelcontextprotocol/sdk

# Or the Python SDK
pip install mcp

An MCP server exposes tools (functions the agent can call), resources (data the agent can read), and prompts (reusable instructions). Most servers only need tools.

See the MCP specification for the full protocol reference and the official server examples for implementation patterns.

Tips

  • Start with one or two servers. Adding too many at once can slow down session startup and increase token usage from tool descriptions.
  • Use autohand mcp list to verify servers are connected before starting a task that depends on them.
  • Set API keys as environment variables rather than hardcoding them in config files.
  • Stdio servers run as local processes and stop when your session ends. HTTP and SSE servers stay available across sessions.
  • If a server fails to connect, check its logs with autohand mcp logs <name> for troubleshooting.