Claude Code MCP Connectors — How to Link GitHub, Sentry, and Databases

Connecting an external tool to Claude Code comes down to running one command to add an MCP (Model Context Protocol — the open standard that lets Claude talk directly to services like GitHub, Notion, or Sentry) connector. Follow this guide and in 15–20 minutes you'll have at least one commonly used connector (GitHub, Sentry, or a database) up and running, plus a sense of which connector to reach for next.

What is an MCP connector?

An MCP connector is what lets Claude Code read and act on an external service — a GitHub issue, a Sentry error log, a Notion doc, a database — directly instead of you copying data into the chat. The official docs frame it simply: connect a server when you find yourself pasting data from another tool into chat, like an issue tracker or a monitoring dashboard. Once connected, you can just ask, "What's the most common error in Sentry from the last 24 hours?" and Claude queries the service itself.

Technically, a connector uses one of two transports. Remote services use HTTP (Claude connects to a server over the internet); tools that need to run on your own machine use stdio (a local process). Each connector's docs tell you which one to use — you don't need to choose, just follow the example command.

What you need before starting

• Claude Code CLI installed and logged in (confirm the claude command runs in your terminal first)
• A terminal — macOS, Windows, or Linux all work
• An account on the service you want to connect (e.g., GitHub), and admin access if you need to issue an access token

Quick glossary

• MCP (Model Context Protocol) — the open standard Claude uses to reach external tools and data
• Connector — a single MCP server linked to a specific service
• Transport — how a connector connects: HTTP for remote services, stdio for local ones
• Scope — whether a connector is private to you, shared with your project team, or available across all your projects
• PAT (Personal Access Token) — a personal credential used instead of a password, issued by services like GitHub
• OAuth — a browser-based sign-in flow used to authenticate securely

Which connectors should you set up first?

A practical order is: GitHub if you're doing code work, Sentry if you're chasing bugs, Notion if you live in docs, and a database connector if you query data often. The table below covers connectors with verified example commands from the official docs.

Step-by-step setup

Setting up an MCP connector comes down to three steps: ①add it with a command → ②authenticate via /mcp → ③use it in conversation. Below are real, verified example commands.

Step 1 — Add the GitHub connector (remote HTTP example)

GitHub's official remote MCP server authenticates with a personal access token (PAT) passed as a header. Generate a fine-grained token scoped to the repositories you want Claude to work with, then add the server:

claude mcp add --transport http github https://api.githubcopilot.com/mcp/ \
  --header "Authorization: Bearer YOUR_GITHUB_PAT"

Watch out: claude mcp add saves the configuration without validating the token, so a bad token is accepted at this step and only fails later. You must check the connection status in the next step.

Step 2 — Authenticate and verify with /mcp

Inside Claude Code, run:

/mcp

For OAuth connectors (Sentry, Notion, etc.), this opens a browser sign-in flow — just follow the prompts. Success looks like: the connector shows as connected in the /mcp list. If credentials are wrong, it shows failed — double-check your token in that case.

Step 3 — A locally-run connector, e.g. a database (stdio example)

PostgreSQL databases connect through a local tool called dbhub:

claude mcp add --transport stdio db -- npx -y @bytebase/dbhub \
  --dsn "postgresql://readonly:[email protected]:5432/analytics"

Watch out: stdio connectors require a -- (double dash) separator. Everything before it is Claude Code's own options (--transport, --env, etc.); everything after is the actual program and its arguments. Omitting -- causes Claude Code to misread the server's own flags as its own options, producing an error.

Once connected, you can ask things like "What's our total revenue this month?" in plain language.

Common problems and fixes

• Added but not actually working — Adding a connector only saves the configuration. You must also run /mcp to authenticate before it actually connects.
• 401 Unauthorized / 403 Forbidden — Usually an expired or under-scoped token. Re-authenticate via /mcp or issue a fresh token.
• Following an old npm-package tutorial gets stuck — Older local packages like @modelcontextprotocol/server-github and @modelcontextprotocol/server-postgres are deprecated and no longer recommended. If you see a guide pointing to these, use the modern remote connector (GitHub's official endpoint) or a tool like dbhub instead.
• A stdio command's later flags get parsed strangely — You likely forgot the -- separator. As in the database example above, place it right before the program you're running (-- npx ...).

Scope — private to you or shared with your team

The -s or --scope flag controls how widely a connector is shared. With no flag, it defaults to local — private to you, in the current project only. project scope stores the config in .mcp.json, which you can commit to version control (e.g., git) to share with teammates, who each approve it on first run. user scope makes it available across all your projects.

# Example: sharing with your team
claude mcp add --transport http paypal --scope project https://mcp.paypal.com/mcp

Managing connectors after setup

Going further

To browse more options, check the Anthropic Directory, which lists reviewed connectors you can add with the same claude mcp add command shown above. If you've already connected services at claude.ai/customize/connectors, they'll appear automatically in Claude Code once you're signed in with the same claude.ai account. For the full reference, see the official MCP documentation.

Frequently asked questions

Q. Does adding a connector make it usable right away?
No. After running claude mcp add, you still need to authenticate with /mcp before it actually connects. The add command doesn't validate credentials, so always confirm the connector shows as connected in /mcp.

Q. How do I share the same connector with my whole team?
Add it with --scope project so it's stored in the project's .mcp.json file, which you can commit to version control. Each teammate still goes through an approval step the first time they run it.

Q. If I already added a connector on claude.ai, does it show up in Claude Code too?
Yes — if you're signed into Claude Code with the same claude.ai account, connectors you added there appear automatically in the /mcp list. This doesn't apply if you're authenticating with an API key or another method.

Q. Should I use an old local npm package or the remote HTTP connector?
Prefer the remote HTTP option where available. It's the most widely supported transport, and some older local packages — like the legacy GitHub and PostgreSQL npm servers — are deprecated and no longer recommended.

Looking for more connectors? Browse the connector directory to search and filter common connectors by use case.

Related articles

• Claude Code MCP Setup: claude mcp add, Scopes, and Transports
• Adding a Custom Connector to Claude Code — Complete MCP Setup Guide
• Connect Claude to Figma: Figma MCP (Dev Mode) Setup for Claude Code (2026)
• Claude Code settings.json Guide — Permissions, Env Vars & Tools
• Claude Connectors & MCP by Category: The Full Catalog (2026)