MCP Integration
Model Context Protocol (MCP) servers extend agents with external tools that go beyond code — browser automation, database access, filesystem operations, web search, and more.
What Is MCP?
MCP is an open protocol that lets agents connect to external tool servers. An MCP server exposes tools, resources, and prompts that an agent can call during a session. Think of it as a plugin system for agent capabilities — but instead of injecting instructions, it injects live tools.
The key difference from skills: skills inject knowledge and workflows. MCP servers inject live tools. They're complementary.
Built-in MCP Servers
oMO ships with three MCP servers always available:
| Server | What It Does |
|---|---|
| Exa (websearch) | Web search with clean, ready-to-use content. Finds current information, news, facts, people, companies. |
| Context7 | Queries up-to-date documentation and code examples for any library or framework. Resolves library IDs, then fetches docs. |
| Grep.app | Searches real-world code across over a million public GitHub repositories. Finds production patterns and usage examples. |
These are always on. No configuration needed.
Skill-Embedded MCPs
The real power: skills can carry their own MCP servers. This is an oMO innovation that solves the context bloat problem.
Without skill-embedded MCPs, every MCP server you connect stays in context for the entire session, consuming your context budget whether you use it or not. With skill-embedded MCPs:
- The MCP server spins up on-demand when the skill is loaded
- It's scoped to the task the skill is handling
- It shuts down when the skill's task is done
- Your context window stays clean
// A skill with an embedded MCP
// When this skill loads, its MCP server starts
// When the skill finishes, the MCP server stops
task(category="visual-engineering", load_skills=["frontend-design"], prompt="...")Connecting External MCP Servers
You can connect any MCP-compatible server to your agent. Common integrations:
Playwright (Browser Automation)
{
"mcps": {
"playwright": {
"command": "npx",
"args": ["@anthropic/mcp-playwright"]
}
}
}Enables browser automation: navigate pages, click elements, fill forms, take screenshots, extract content. Useful for E2E testing, web scraping, and visual verification.
Filesystem (Extended File Access)
{
"mcps": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/dir"]
}
}
}Grants the agent access to directories outside the project root. Useful for cross-project operations, accessing shared assets, or working with system files.
Database (Direct Query Access)
{
"mcps": {
"sqlite": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-sqlite", "--db-path", "/path/to/db.sqlite"]
}
}
}Lets the agent query databases directly. Useful for debugging data issues, verifying migrations, and exploring schemas.
GitHub (Extended Repository Operations)
{
"mcps": {
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": {
"GITHUB_PERSONAL_ACCESS_TOKEN": "${GITHUB_TOKEN}"
}
}
}
}Extends beyond the built-in gh CLI with richer repository operations, issue management, and PR workflows.
MCP vs Skills
| Aspect | Skills | MCP Servers |
|---|---|---|
| What they add | Knowledge, workflows, instructions | Live tools and data access |
| Context cost | Loaded on-demand, released after task | Persistent while connected |
| Scope | Domain-specific (TDD, debugging, etc.) | Tool-specific (browser, DB, etc.) |
| Composition | Skills can embed MCP servers | MCP servers can be standalone |
| Creation | Write a SKILL.md | Implement the MCP protocol |
They compose: a skill like frontend-design can embed a Playwright MCP server for visual verification, loading it only when the skill is active and releasing it when done.
Disabling MCP Servers
To disable a built-in MCP server:
{
"disabled_mcps": ["playwright", "browser"]
}This prevents the server from starting, even if it's referenced by a skill. Use this when a server conflicts with your environment or you don't need its capabilities.