MCP Server Catalog

What is MCP, and where do I start?

MCP (Model Context Protocol) is Anthropic's open standard for AI agents to connect to external tools and data. An MCP server exposes a service to Claude as a "tool". Claude can call it; the result enters context and informs the next reply.

Servers are proliferating; the right one depends on your workflow. This guide categorizes the common ones, with a typical install pattern and the most common pitfall for each.

Most servers run via npx or uvx. Configure them in claude_desktop_config.json for Claude Desktop, or ~/.claude/mcp.json / project .mcp.json for Claude Code.

Anatomy of a config

Every server takes three things: command (how to run), args (parameters), env (secrets). Minimal:

{
  "mcpServers": {
    "<name>": {
      "command": "npx",
      "args": ["-y", "<package>", "<arg1>"],
      "env": {
        "SECRET_TOKEN": "${SECRET_TOKEN}"
      }
    }
  }
}

Use ${VAR} placeholders and keep secrets in .env so configs are safe to commit.

Development environment

filesystem

Local read/write access. Claude opens and edits files directly.

"filesystem": {
  "command": "npx",
  "args": [
    "-y",
    "@modelcontextprotocol/server-filesystem",
    "/Users/me/projects",
    "/Users/me/Documents/notes"
  ]
}

Pitfall: never set the root to / or your home folder — that hands Claude every file. Restrict to project subdirectories.

git / github

git for local commit/branch ops; github for issues, PRs, branch info.

"github": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"],
  "env": { "GITHUB_TOKEN": "${GITHUB_TOKEN}" }
}

Pitfall: scope the token tightly. repo:read is usually enough — never expose delete_repo etc.

sentry

Query production errors. Great for "top 5 errors in the last 24h".

"sentry": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-sentry"],
  "env": { "SENTRY_AUTH_TOKEN": "${SENTRY_AUTH_TOKEN}" }
}

Databases

postgres

Run queries, read schema. Connect with a read-only user.

"postgres": {
  "command": "npx",
  "args": [
    "-y",
    "@modelcontextprotocol/server-postgres",
    "postgresql://readonly_user:pass@host:5432/mydb"
  ]
}

Pitfall: connecting with a read-write user is a foot-gun. Provision a dedicated user: CREATE USER claude_ro WITH LOGIN; GRANT SELECT ON ALL TABLES IN SCHEMA public TO claude_ro;.

sqlite

Local dev databases, small analytics datasets.

"sqlite": {
  "command": "uvx",
  "args": ["mcp-server-sqlite", "--db-path", "/path/to/db.sqlite"]
}

mongodb / mysql / redis

Mature community packages; check awesome-mcp-servers for the latest names.

Communication

slack

Send messages, read channels, look up users.

"slack": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-slack"],
  "env": {
    "SLACK_BOT_TOKEN": "${SLACK_BOT_TOKEN}",
    "SLACK_TEAM_ID": "T01ABC..."
  }
}

Pitfall: if the bot has "post to all channels", a single mistake spams the company. Restrict scope.

gmail / google-drive

Use OAuth; first run pops a browser for consent.

"gdrive": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-gdrive"]
}

linear / notion / jira

Issue-tracker integration — "list my open tickets", "comment on this".

Cloud & SaaS

stripe

Look up customers, payments, subscriptions. Use carefully in production.

aws-s3

List bucket contents, download/upload files. IAM least-privilege policy.

cloudflare

DNS records, domain management, worker deploys.

asana / trello

Lightweight project management integration.

Web and search

brave-search

General web search. Privacy-friendly alternative to Perplexity.

"brave-search": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-brave-search"],
  "env": { "BRAVE_API_KEY": "${BRAVE_API_KEY}" }
}

fetch

Fetches a URL and returns Markdown. Core for any web research workflow.

puppeteer

Full browser automation — page navigation, form filling, screenshots. The only way for JS-heavy sites.

"puppeteer": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-puppeteer"]
}

Pitfall: Puppeteer is powerful, easy to misuse. Sandbox or run in Docker.

AI ecosystem

vector databases

Separate servers for Pinecone, Qdrant, Chroma — useful in RAG dev loops.

"qdrant": {
  "command": "uvx",
  "args": ["mcp-server-qdrant"],
  "env": { "QDRANT_URL": "${QDRANT_URL}", "QDRANT_API_KEY": "${QDRANT_API_KEY}" }
}

huggingface

Search models on HF Hub, download datasets, query model metadata.

Systems and devops

docker

List containers, stop, log. Automates dev workflows.

kubernetes

Pod, deployment, service management. Always read-only in production.

ssh

Run commands on remote hosts. One of the most dangerous MCP servers; never connect to production, restrict to lab hosts.

Five rules for safe use

1. Least privilege: every token, user, and IAM role is granted the minimum scope.
2. Read-only first: experiment with read-only versions; add write later.
3. Sandbox: especially for community servers, run in Docker; restrict network.
4. Audit logs: track which tool Claude called when. Disable servers you don't actually need.
5. Use environment variables: don't bake tokens into committed configs. ${VAR} + .env is the standard.

A common starter set

For most software teams:

{
  "mcpServers": {
    "filesystem":   { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/projects"] },
    "github":       { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-github"],
                      "env": { "GITHUB_TOKEN": "${GITHUB_TOKEN}" } },
    "postgres":     { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-postgres",
                                                  "postgresql://readonly:pass@localhost/mydb"] },
    "brave-search": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-brave-search"],
                      "env": { "BRAVE_API_KEY": "${BRAVE_API_KEY}" } },
    "fetch":        { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-fetch"] }
  }
}

These five cover ~80% of "codebase + external search + DB query + URL fetch" needs. Add more as you go.

Community resources

• Official reference servers: github.com/modelcontextprotocol/servers
• awesome-mcp-servers: github.com/punkpeye/awesome-mcp-servers
• modelcontextprotocol.io: the standard's home page and docs

Continue reading

• MCP — the protocol itself: how an agent talks to the outside world.
• Claude Skills Guide — MCP's natural companion for daily workflows.
• Function Calling — the underlying tool-use behavior MCP relies on.
• AI Agent — who actually uses MCP servers, and how.