Guide
OpenClaw + MCP: Model Context Protocol Skills
How OpenClaw uses MCP to give your AI agent real-world capabilities through 5,700+ skills
What Is the Model Context Protocol (MCP)?
The Model Context Protocol (MCP) is an open standard created by Anthropic that provides a standardized way for AI models to interact with external tools, databases, and APIs. Think of it as USB-C for AI — one universal protocol to connect any tool to any AI model.
Before MCP, every AI integration required custom code. Each tool, each API, each database needed its own connector built from scratch. MCP changes this by defining a single protocol that any AI application can use to discover and call tools — regardless of which AI model powers it.
An MCP server is a lightweight program that exposes a set of tools (called "capabilities") over the MCP protocol. The AI model connects to MCP servers as a client, discovers what tools are available, and calls them during conversations to take real-world actions.
How OpenClaw Uses MCP
OpenClaw is one of the largest MCP-compatible platforms. Every skill on ClawHub — OpenClaw's skill marketplace — is an MCP server. When you enable a skill, OpenClaw connects to that MCP server and makes its tools available to your AI agent.
Each skill exposes one or more tools that the agent can call during conversations. For example, the web-search skill exposes a search tool that the agent calls when you ask it to look something up. The image-gen skill exposes a generation tool that creates images from text prompts.
This architecture means your OpenClaw agent isn't limited to what the AI model knows — it can take action in the real world through any MCP-compatible tool. And because MCP is an open standard, new skills are being built by the community every day.
Popular MCP Skills for OpenClaw
ClawHub hosts over 5,700 MCP skills across dozens of categories. Here are the most popular ones:
Web & Search
Skills: web-search</code>, <code>browser
Search the web, scrape pages, and extract information from any URL.
Coding
Skills: code-runner</code>, <code>git
Execute code in sandboxed environments, manage git repos, and review pull requests.
Files
Skills: file-manager</code>, <code>pdf-reader
Upload, download, and manage files. Read and summarize PDFs, spreadsheets, and documents.
Productivity
Skills: calendar</code>, <code>email-drafter
Schedule events, draft emails, manage to-do lists, and set reminders.
Automation
Skills: http-request</code>, <code>cron-scheduler
Make HTTP API calls, schedule recurring tasks, and build automated workflows.
Image
Skills: image-gen</code>, <code>screenshot
Generate images with AI, capture screenshots, and process visual content.
These are just a fraction of what's available. ClawHub skills cover everything from flight check-ins to workout logging to database queries — all built on the MCP standard.
How to Install MCP Skills
There are two ways to add MCP skills to your OpenClaw agent, depending on how you deploy:
On OpenClaw Launch (managed)
- Open the visual configurator — Go to your dashboard and click on your agent's configuration.
- Browse the Skills section — Scroll through available MCP skills organized by category.
- Toggle skills on — Each skill has a switch. Turn it on and the MCP server is automatically connected to your agent.
- Save and deploy — Your agent immediately starts using the new skills. No restart required — OpenClaw hot-reloads skill changes.
Self-hosted (Docker / CLI)
If you run OpenClaw yourself, add MCP skills by editing the plugins section of your openclaw.json configuration file:
{
"plugins": {
"entries": {
"web-search": { "enabled": true },
"image-gen": { "enabled": true },
"file-manager": { "enabled": true }
}
}
}After saving the config file, OpenClaw's file watcher detects the change and hot-reloads the new skills within seconds.
Building Custom MCP Skills
Because MCP is an open standard, developers can build custom MCP servers that OpenClaw can use. If you need a tool that doesn't exist on ClawHub — say, connecting to your company's internal API or a niche service — you can build your own MCP server and plug it into OpenClaw.
MCP servers can be written in any language that supports the protocol. The MCP specification provides SDKs for TypeScript, Python, and other languages to make development straightforward.
MCP on OpenClaw Launch
OpenClaw Launch makes MCP skills effortless. Instead of manually editing config files and managing MCP server processes, you get:
- Visual skills browser — Browse, search, and filter 5,700+ MCP skills from a graphical interface.
- One-click install — Toggle any skill on or off. No config files, no command line, no Docker knowledge needed.
- 5,700+ pre-configured skills — Every MCP server on ClawHub is pre-configured and ready to use. No setup, no API key management (for skills that don't require external accounts).
- Hot-reload — Skills activate instantly without restarting your agent. Add or remove capabilities in real time.
MCP vs. Traditional Plugin Systems
| MCP (OpenClaw) | ChatGPT Plugins | Custom Integrations | |
|---|---|---|---|
| Open standard | Yes (Anthropic's MCP) | No (proprietary) | No |
| Available tools | 5,700+ on ClawHub | Limited GPTs | Build your own |
| Model support | Any (Claude, GPT, Gemini, etc.) | GPT only | Varies |
| Custom tools | Yes (build MCP servers) | Limited | Yes |
| Hot-reload | Yes | No | Usually no |
| Self-host option | Yes | No | Yes |