Girish KotteBook a call
AI kit · MCP guide

MCP servers, connect AI to everything.

MCP (Model Context Protocol) is an open standard that lets AI models like Claude connect to external tools, APIs, and data sources. Build integrations that extend what large language models can do.

OpenStandard by Anthropic
PythonAnd TypeScript SDKs
3Tools, resources, prompts
ClaudeNative support
Add a server in one lineOpen standard
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/Documents
# then, in Claude Code
> list the PDFs in my Documents folder
01ClientClaude, IDEs
02ServerExposes tools
03ToolsActions
04ResourcesData to read
In Claude Desktop, add the same server to claude_desktop_config.json (section 04).
01

What is MCP?

An open standard created by Anthropic that lets AI assistants connect to external data sources and tools in one consistent way.

Three things to know
Universal standardOne protocol to connect AI to any tool or data source
Developer friendlySDKs for Python, TypeScript, and more languages
Growing ecosystemReference, official, and community servers to start from
Perfect forFile system accessDatabase queriesAPI integrationWeb browsingCustom tools
What a server can expose
ToolsFunctions the AI can call: read and write files, run queries, call APIs, run commands
ResourcesData the AI can read: documents, database records, API responses
PromptsTemplates that show the AI how to use a server well
Clear permissionsClients ask before tools run, and local servers only reach what you allow
02

How MCP works

Three parts: the app you talk to, the server that exposes capabilities, and the data or actions behind it.

1. MCP client

Apps like Claude Desktop, Claude Code, or an IDE that want to use outside capabilities

2. MCP server

A lightweight service that exposes specific capabilities through the standard protocol, locally or remotely

3. Tools and resources

The data, APIs, or functions the server makes available to the AI

03

Server directory

Servers worth knowing, with where each one lives today. Search or filter by category.

24 of 24 shown

FFilesystemReferenceRead, write, and manage files on your machine, limited to the folders you allowReference · MCP teamFFetchReferenceFetch web pages and turn them into text the model can readReference · MCP teamGGitReferenceRead, search, and work with local Git repositoriesReference · MCP teamMMemoryReferenceA knowledge graph the model can use as long-term memoryReference · MCP teamGGitHubOfficialInteract with GitHub repositories, issues, pull requests, and moreDevelopment · GitHubPPostgreSQLConnect to PostgreSQL databases for read-only queries and schema explorationData · Archived referenceSSlackRead and send messages, manage channels, and interact with Slack workspacesCommunication · Archived referenceGGoogle DriveSearch and read files in Google DriveProductivity · Archived referencePPuppeteerWeb browser automation for scraping, testing, and interacting with websitesWeb · Archived referenceNNotionRead and write Notion pages, databases, and workspace contentProductivity · CommunityLLinearManage Linear issues, projects, and team workflowsProductivity · CommunityGGoogle CalendarCreate, read, and manage Google Calendar eventsProductivity · CommunityTTodoistManage tasks, projects, and labels in TodoistProductivity · CommunityDDockerManage Docker containers, images, and compose stacksDevelopment · CommunityAAWSInteract with AWS services including S3, EC2, Lambda, and moreDevelopment · CommunityKKubernetesManage Kubernetes clusters, deployments, and resourcesDevelopment · CommunityGGitLabInteract with GitLab repositories, CI/CD, and merge requestsDevelopment · CommunitySSQLiteQuery and manage SQLite databases locallyData · CommunityMMongoDBConnect to MongoDB for document queries and aggregationsData · CommunityBBigQueryRun queries on Google BigQuery datasetsData · CommunityDDiscordSend messages and interact with Discord serversCommunication · CommunityEEmail (IMAP)Read and search emails from IMAP-compatible email serversCommunication · CommunityBBrave SearchSearch the web using the Brave Search APIWeb · CommunityEExaNeural search for finding relevant content across the webWeb · Community
Several early Anthropic servers (PostgreSQL, Slack, Google Drive, Puppeteer) moved to the archived repository, and GitHub now maintains its own server. Built one? Submit it to the official list at github.com/modelcontextprotocol/servers.
04

Build your own MCP server

Create a custom server to connect Claude to your own tools and data. Install the SDK, define tools, add resources, then add it to your client config.

# pip install "mcp[cli]"
from mcp.server.fastmcp import FastMCP

# Create server instance
mcp = FastMCP("my-mcp-server")

# Define a tool
@mcp.tool()
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    # Your implementation here
    return f"Weather in {city}: Sunny, 72°F"

# Define a resource
@mcp.resource("config://settings")
def get_settings() -> str:
    """Return application settings."""
    return '{"theme": "dark", "language": "en"}'

# Run the server (stdio)
if __name__ == "__main__":
    mcp.run()
05

What can you build with MCP?

Popular setups, the servers they use, and a prompt to try.

AI-powered code review

GitHub, Filesystem

Connect Claude to your GitHub repository to review pull requests, suggest improvements, and check for security issues.

Review the latest PR and check for potential bugs

Database assistant

PostgreSQL, SQLite

Let Claude query your databases, find patterns, and write reports without handing it raw credentials.

Analyze sales trends from the last quarter

Automated documentation

Filesystem, GitHub

Have Claude read your codebase and write or update the docs.

Document all API endpoints in this project

DevOps automation

AWS, Docker, Kubernetes

Manage infrastructure, deploy applications, and troubleshoot issues with AI help.

Check the status of the production cluster

Smart calendar management

Google Calendar, Slack

Let Claude schedule meetings and find open times from plain requests.

Find a time for a team meeting next week

Research assistant

Brave Search, Fetch

Search the web, gather information, and pull findings into one report.

Research competitor pricing strategies
FAQ

Questions,
answered.

The six questions people ask most.

Ask on a call

MCP is an open standard created by Anthropic that lets AI assistants like Claude connect to external tools, data sources, and APIs. It gives AI apps one standard way to work with the outside world.

Most servers run with npx or uvx, so there is nothing to install by hand. Add the server to your client's config (for Claude Desktop, claude_desktop_config.json) and restart the app. Remote servers connect with a URL instead.

Local servers run on your machine and only reach what you give them, such as the folders you allow. Clients like Claude ask before a tool runs. Remote servers send data to their host, so only connect servers you trust.

Claude (desktop, web and Claude Code) supports it natively, and so do many other tools, including ChatGPT, Cursor, VS Code with GitHub Copilot, and Windsurf. The protocol is open, so any app can add support.

Tools are functions the AI can call to take actions, like writing a file or calling an API. Resources are data the AI can read, like file contents or database records. Servers can expose both, plus prompts.

MCP itself is free and open source. Claude Desktop supports local MCP servers on the free plan; some servers need paid accounts with the service they connect to.

Next step

Ready to extend Claude with MCP?

Connect Claude to your tools and data with the quickstart. If you want MCP built into your product, book a call.