If you're not sure which MCP server to connect to Claude Code, the answer isn't a specific product name — it's whatever information you currently copy out of another tool and paste into the chat. MCP (Model Context Protocol) — an open standard that lets AI tools like Claude connect to external tools and data lets Claude Code read and act on those tools directly. This guide walks through what's worth connecting first, how to actually add a server, and where people commonly get stuck, based on the official documentation, in about 15-20 minutes.
🟢 Model references match the current lineup · model notice · Fable subscription
Fable 5 and 5.1 subscription (updated September 7, 2026): Claude Fable 5.1, released September 1, 2026, is the current Fable model and Fable 5 is now legacy. Plan terms are the same for both — Max and Team Premium plans include Fable at up to 50% of the weekly usage limit; Pro and Team Standard use usage credits (
What you need before starting
Make sure you have the following so you can follow along.
- Claude Code CLI installed and logged in
- A terminal where you can run commands
- An account for the external service you want to connect (an issue tracker, database, monitoring tool, etc.)
- A browser you can log into, since some services require OAuth authentication during setup
Quick terms
- MCP (Model Context Protocol) — an open standard that lets AI connect to external tools and data
- MCP server — a program following this standard that bridges a specific tool (GitHub, a database, etc.) and Claude Code
- Remote server — an MCP server that runs in the cloud; you connect with just an address, no local install
- Local stdio server — an MCP server that runs directly on your own machine
- Transport — the communication method Claude Code and an MCP server use (HTTP, SSE, and others)
- Scope — where an added server's configuration is stored: local, project, or user level
Which category of MCP server should you connect first?
The highest-value pick is whatever category replaces work you currently do by copying information from another screen into Claude Code. The official docs give these examples.
| Category | Official example | What it lets you do |
|---|---|---|
| Issue tracker | JIRA | Reference an issue number and have Claude read it, implement the feature, and open a PR |
| Monitoring/analytics | Sentry, Statsig | Pull error logs or feature usage data to diagnose issues |
| Database | PostgreSQL | Query the database directly for the data you need |
| Design | Figma | Reference design files while implementing a screen |
| Communication/automation | Slack, Gmail, Telegram, Discord, webhooks | Read messages, draft and send replies, or get pushed a message when an external event fires |
Among these, the official docs recommend remote HTTP servers as the most widely supported option for cloud-based services. Start by connecting just one, then expand to other categories once you're comfortable.
Steps to add an MCP server to Claude Code
-
Pick one tool to connect. Use the category table above to choose whichever tool you rely on most.
-
Find a reviewed server in the Anthropic Directory. Directory connectors use the same MCP infrastructure as Claude Code, so you can add any listed server with
claude mcp add. Caution: always verify you trust a server before connecting it. Servers that fetch external content can expose you to prompt injection — hidden instructions embedded in fetched content that trick the AI into unintended actions. -
Run the add command in your terminal. Remote HTTP is the most widely supported transport per the official docs, and the basic syntax is:
claude mcp add --transport http <name> <url>SSE, local stdio, and WebSocket transports are also supported, but exact flag syntax can vary by server — follow the exact command shown in that server's own setup instructions (official docs or its Directory listing).
-
Complete authentication if prompted. Many remote servers require OAuth authentication — a login method that grants permission without sharing your password. Follow the on-screen prompts after running the command. Once connected, Claude Code can read and act on that tool's data in later conversations. Check the exact confirmation label in your current Claude Code screen, since UI text can change between versions.
Common issues and fixes
I followed old instructions for packages like @modelcontextprotocol/server-github or @modelcontextprotocol/server-postgres and it doesn't work. Those packages are now deprecated and not recommended for new connections. Use the Anthropic Directory to find a currently maintained remote server and connect it with claude mcp add instead.
Is it safe to connect any MCP server I find? No. Servers that fetch external content carry prompt injection risk, so always verify the provider and trustworthiness of a server before connecting. Avoid servers from unclear or unverified sources.
How do I limit permissions in real projects? Two habits help. First, least privilege — connect database servers (such as PostgreSQL) with a read-only account, and explicitly narrow the allowed directories for filesystem servers. Second, project-scoped configuration — instead of a global setup, enable only the servers each project needs via a project-level .mcp.json, which naturally isolates permissions between projects.
Authentication keeps failing. Check whether the browser login window is opening correctly and whether a company network or firewall is blocking the authentication request. The official docs cover several alternatives, including authenticating from the command line and using a fixed OAuth callback port — check which applies to your setup.
Next step: building your own MCP server
If the tool you want isn't in the Directory, Claude Code has an official plugin that helps you scaffold one.
- Install the plugin: run
/plugin install mcp-server-dev@claude-plugins-official. If you see "Marketplace not found," add it first with/plugin marketplace add anthropics/claude-plugins-official; if the plugin still isn't found, refresh with/plugin marketplace update claude-plugins-officialand retry. - Activate it in your session: run
/reload-pluginsafter installing. - Start building: run
/mcp-server-dev:build-mcp-serverand Claude will ask about your use case, then scaffold either a remote HTTP or local stdio server.
Frequently asked questions
Q. Do I have to use only servers listed in the Directory?
No, you can connect any server regardless of whether it's listed, but unverified servers carry prompt injection risk, so you need to confirm trustworthiness yourself. If you're just starting out, sticking to Directory-listed servers is the safer choice.
Q. Can I still use older npm packages like @modelcontextprotocol/server-github?
It's not recommended. Those packages are deprecated, so it's better to find a currently maintained remote server in the Directory instead.
Q. Should I choose a local or remote server?
For cloud-based services, remote HTTP is the most widely supported option per the official docs. A local stdio server may be needed if the tool is only accessible from your own machine.
Q. Do I need to re-add the same MCP server in every project?
No. Scope settings — local, project, or user — let you control where a server applies. Check the scope section of the Claude Code docs for the exact setup for your case.
Looking for more connectors? Browse the connector directory to search and filter common connectors by use case.
Related articles
- Claude Code MCP Connectors — How to Link GitHub, Sentry, and Databases
- Claude Code settings.json Guide — Permissions, Env Vars & Tools
- PostgreSQL MCP Server: Connect a Database to Claude (2026)
- Build Your Own MCP Server: From First Python Server to Claude (2026)
- Playwright MCP Server Setup: Browser Automation in Claude (2026)
- Claude Code MCP Setup: claude mcp add, Scopes, and Transports
If you want image and video generation from Claude, see using ComfyUI from Claude (Comfy MCP) — it covers the no-install cloud path and what it costs.