Docs/MCP Servers

MCP Servers

Extend your agent with Model Context Protocol servers for databases, APIs, filesystems, and more.

What is MCP?

Model Context Protocol (MCP) is an open standard developed by Anthropic that lets AI agents connect to external tools, databases, and services through a unified interface.

Instead of hardcoding integrations, MCP servers expose resources (files, databases, API endpoints) and tools (functions your agent can call) in a standard way. Your agent uses these automatically.

🔌

One Protocol, Many Integrations

MCP servers are language-agnostic (Python, TypeScript, Go, etc.) and work across any MCP-compatible AI system. Pinchr natively supports MCP out of the box.

How MCP Works in Pinchr

When you add an MCP server to Pinchr, your agent gains access to:

📁

Resources

Files, databases, documents — things your agent can read and search through.

🛠️

Tools

Functions your agent can invoke: run queries, create records, call APIs, trigger webhooks.

You:

"Check our GitHub issues and find all P0 bugs."

Agent:

I'll use the GitHub MCP server to search issues...

→ Agent calls github_search_issues(label:"P0", state:"open")

Agent:

Found 3 P0 bugs: [list with details]

Everything happens automatically. You don't manually configure tools — your agent figures it out.

Adding MCP Servers

Pinchr makes it easy to add MCP servers. Go to Settings → MCP Serversand choose from presets or add a custom server.

✨

Agent-Assisted Setup

Your agent helps you configure MCP servers. Just tell it what you want to connect: "Connect my GitHub repos" or "Add Supabase access" — and it handles the rest.

Once added, the server runs in the background. Your agent automatically discovers and uses available tools and resources.

Available MCP Presets

Pinchr includes pre-configured MCP servers for common services:

🐙

GitHub

Search repos, read code, create issues, manage PRs, run workflows. Full GitHub API access.

🗄️

Supabase

Query Postgres tables, insert/update records, run functions. Direct database access.

📂

Filesystem

Read/write local files, search directories, watch for changes. Sandboxed to specific folders.

📊

Google Sheets

Read and update spreadsheets, run formulas, export data. Full Sheets API access.

📧

Gmail

Search emails, read threads, send messages, manage labels. Full Gmail integration.

📝

Notion

Search pages, read databases, create/update content. Full workspace access.

💬

Slack

Read channels, send messages, search history, manage threads. Full Slack API.

☁️

AWS

Manage S3 buckets, EC2 instances, Lambda functions, and more. Full AWS SDK access.

More presets coming regularly. You can also add custom MCP servers from the community.

Adding Custom MCP Servers

You can connect any MCP-compatible server. In Settings → MCP Servers, click Add Custom Server and provide:

Server Name

Internal API

Command

node /path/to/mcp-server.js

Environment Variables (optional)

API_KEY=your_key_here

Pinchr launches the server automatically and maintains the connection. Your agent discovers available tools on startup.

💡

Find community MCP servers at github.com/topics/mcp-server. Popular servers include Stripe, Twilio, Redis, and hundreds more.

Security & Permissions

MCP servers run in isolated processes. Pinchr manages access control:

•Tool calls are visible: You see every MCP tool your agent uses in the audit log
•Sandboxed by default: Filesystem access is scoped to specific directories
•Revocable: Disable or remove servers anytime from Settings
•API keys stay local: Never sent to Anthropic or OpenAI servers

⚠️

Review custom MCP server code before adding. MCP servers can execute arbitrary code. Only use servers from trusted sources or your own infrastructure.

Building Your Own MCP Server

Developers can build custom MCP servers for internal tools, APIs, or databases. The MCP SDK supports Python, TypeScript, and Go:

// TypeScript example

import { Server } from '@anthropic-ai/mcp-server'

const server = new Server()

server.tool('get_user', async (args) => {

return await fetchUser(args.userId)

})

Full documentation for building MCP servers: modelcontextprotocol.io