Model Context Protocol

MCP: AI That Does Things

Without MCP, Claude knows only what you type. With MCP, Claude can open your Google Drive, send a Slack message, create a GitHub issue, and query your database — in real time, without you doing it manually. This is how AI goes from chatbot to agent.

The Concept

What MCP Actually Is

The short version: MCP is the bridge between Claude and the rest of your tools.

The contractor analogy

Without MCP, Claude is a contractor who only works with whatever you hand them. You have to copy-paste files, describe problems, relay information back and forth. Claude can only see what you put in the chat window.

With MCP, Claude can act

With MCP, Claude can go open the filing cabinet itself. Check the calendar. Update the spreadsheet. Send the Slack message. Pull the GitHub PR. It still asks for your approval on anything consequential — but it stops waiting for you to be the middleman.

A universal standard

MCP is an open protocol developed by Anthropic. It is not proprietary to Claude — any AI agent can implement it. That means the integrations you build with MCP today work across the agent ecosystem, not just with one vendor.

Why This Changes Everything

The gap between "AI that knows things" and "AI that does things" is MCP. Before it, using AI on real work meant manually bridging every tool. Copy this from Drive, paste it into Claude, take Claude's response, put it into Slack. You were the API.

MCP removes you from that loop. Claude connects directly to the tools your work actually lives in. That is the difference between a very smart assistant and an agent that actually gets things done.

Available Connectors

What You Can Connect Claude To

These are the MCP connectors available via Claude.ai desktop and Claude Code as of 2026. Authentication is handled once through Claude settings — no manual API key management for most of them.

01

Google Drive

Claude can read, write, and search files in your Drive. Drop a brief in Drive and tell Claude to pull it, summarize it, and turn it into a structured build plan — without you copying anything. Useful for teams where all project docs live in shared drives.

What Claude can do: List files, read document content, search across Drive, create new documents, update existing files.

02

Gmail

Claude can read your inbox, draft replies, and send emails on your behalf. Give Claude context on a situation and ask it to draft a professional follow-up — it reads the thread, writes the response, and sends it. Or reads your emails to pull action items automatically.

What Claude can do: Read emails and threads, search inbox, draft and send emails, create drafts for review.

03

Google Calendar

Claude can read your schedule and create events. "Find me a 30-minute slot this week with John and block it for a design review" — Claude checks your calendar, finds the gap, and creates the event. No more tab-switching to check availability while mid-conversation.

What Claude can do: Read events and free/busy status, create and update events, search for scheduling gaps.

04

Slack

Claude can read channels and send messages. Point Claude at a Slack thread and ask it to summarize what was decided. Or have Claude post a build update to your #deploys channel after a successful Firebase release. Useful for teams that live in Slack.

What Claude can do: Read messages and threads, search channels, send messages to channels or DMs.

05

GitHub

Claude can read your repos, open issues, create PRs, and review code. Ask Claude to look at the open issues on a repo and suggest which ones are blocking the next release. Or have it create a properly-formatted PR with description filled in from the commit history.

What Claude can do: Read repos, branches, and commits; read and create issues; read and create PRs; read and comment on code.

06

Figma

Claude can read and interact with Figma design files. Ask Claude to extract design tokens from a Figma component, describe the layout of a screen, or check whether a design matches your existing style system. Useful for bridging design and development conversations.

What Claude can do: Read file structure and component names, extract design properties, describe layouts.

07

Asana

Claude can create and manage Asana tasks. When you finish planning a build in conversation with Claude, have it write all the tasks directly to Asana with due dates and assignments — no manual re-entry from your planning doc to your project manager.

What Claude can do: Read projects and tasks, create new tasks, update task status and assignees.

08

PostgreSQL

Claude can query your database directly. Point it at your Postgres instance and ask it to explain the schema, write a query, or debug a slow lookup. The query runs against your real data — not a description of your data. This is genuinely powerful for data analysis work.

What Claude can do: Read schema and table structure, run read queries, explain query performance, write and test SQL.

Safety Note Claude will not run destructive queries (DROP, DELETE, UPDATE) without explicit confirmation. Set your MCP connection to a read-only user for production databases to be safe.
09

Brave Search

Real-time web search inside Claude. Instead of Claude relying only on its training data (cutoff August 2025), Brave Search gives it live access to current information. Ask Claude to research a competitor, check the latest pricing on a tool, or find recent documentation updates — it actually searches.

What Claude can do: Search the web in real time, return cited results, research recent events and pricing.

10

Custom MCP Servers

You can build your own MCP server that exposes any internal tool or API to Claude. Your internal CRM, your proprietary database, your company's custom API — if you can write a server, you can expose it to Claude via MCP. The protocol is open-source and documented at modelcontextprotocol.io.

# Install a skill that uses MCP via Claude Code
npx skills add owner/repo/skill-name

# Example
npx skills add anthropics/skills
Getting Started

How to Connect Your First MCP Tool

No API keys. No config files. Claude handles the auth flow.

01

Open Claude settings

In Claude.ai desktop app: click your profile icon → Settings → Integrations. In Claude Code: run /mcp in the terminal to see connected servers and add new ones.

02

Choose a connector

Pick from the available integrations list. Google Drive, Gmail, Calendar, Slack, GitHub, and others are listed here. Click the connector you want to add.

03

Authenticate once

Claude will open the OAuth flow for the service. Sign in with your account (Google, GitHub, Slack, etc.). Grant the permissions it requests. Claude stores the connection — you do not re-authenticate on every conversation.

04

Start using it

The tool is now available in your Claude conversations. You do not need to explicitly invoke it — Claude will use it when relevant, or you can ask it directly: "Read my last 5 emails" or "Check my GitHub issues for the alchemy-ai repo." Claude will confirm before taking any write actions.

Real World Use Cases

What This Looks Like in Practice

The solo builder workflow

Claude reads the project brief from Drive. Creates GitHub issues for each build step. Opens VS Code via Claude Code and starts building. Posts a status update to Slack when each milestone is done. You said "build the auth flow" once. Claude did seven things.

The research workflow

Claude searches Brave for the latest competitor pricing. Reads three articles. Pulls the existing research doc from Drive. Writes an updated competitive analysis and saves it back to Drive. What used to take two hours took five minutes.

The data analysis workflow

Claude connects to your Postgres database. You ask: "Why are signups dropping on mobile?" Claude queries the events table, spots the funnel breakage at step 3, writes the SQL that confirms it, and drafts the summary for your team. No BI tool. No manual export.

Ready to Build With MCP?

Start with the CLAUDE.md workflow, then layer MCP on top once your project structure is solid.