What is the Model Context Protocol?
The Model Context Protocol (MCP) is an open standard developed by Anthropic that defines how AI models connect to external data sources and tools. An MCP server exposes a set of tools that an AI agent can call during a conversation — for example, a GitHub MCP server exposes tools like create_pull_request, list_commits, and read_file. The agent decides when to call these tools based on the task at hand.
Why use MCP servers with AI agents?
Without MCP, AI agents are limited to knowledge from their training data. With MCP servers, they can read live data from GitHub repositories, search the real-time web, create and update Notion pages, query databases, post Slack messages, and much more. MCP turns a conversational model into an agent that can actually take actions in the world.
Available MCP servers
Popular MCP servers include: Brave Search (real-time web search with privacy), GitHub (repository access, PR creation, code review), Notion (read/write pages and databases), Slack (send messages, read channels), PostgreSQL and SQLite (database queries), Filesystem (read/write local or mounted files), Puppeteer (browser automation for web scraping), and many more available in the community registry.
Traditional MCP server setup (the hard way)
In a traditional self-hosted Claude Code or AI agent setup, adding an MCP server requires editing a JSON configuration file, specifying the server command and arguments, managing environment variables for API tokens, and restarting the agent. Each new server is another manual JSON edit. Debugging connection issues requires reading server logs directly.
MCP server setup in OpenClaw (the easy way)
OpenClaw provides a visual MCP server manager. In the MCP tab of your dashboard, you see all available servers organized by category (Development, Research, Communication, Database, Finance, DevOps, Design). To connect a server: click Connect, fill in any required credentials (API token, database URL, etc.) in the UI form, and click Save. The agent can use the server's tools immediately. No JSON files, no terminal, no restarts.
Assigning MCP servers per agent
In OpenClaw, MCP servers are assigned per agent rather than globally. This means your Developer agent (Rex) can have GitHub and filesystem MCP access, while your Researcher agent (Nova) has Brave Search and Notion. This prevents agents from having access to tools irrelevant to their role, reducing hallucination risk and keeping costs down.
Getting started with MCP in OpenClaw
To set up MCP servers on OpenClaw: launch your dashboard at open-claw-setup.com (takes 60 seconds), open the MCP tab, and click Connect on the servers you want. For GitHub, you'll need a Personal Access Token. For Brave Search, a free API key. For Notion, OAuth authorization. All credentials are encrypted and stored in your account.
Ready to put this into practice?
Claw gives you a full AI team that handles this kind of work automatically.