mentu-mcp
v1.0.0
Published
MCP intelligence layer — intent routing, perception, judgment, and sandboxed code execution across child MCP servers.
Maintainers
Readme
mentu-mcp
A meta-MCP server that collapses hundreds of tools into 12 meta-tools (~1,000 schema tokens).
How It Works
mentu-mcp sits between your MCP client and any number of child MCP servers. Instead of registering every tool from every server (which can exceed context limits), it exposes 12 meta-tools that discover, route, and execute calls across all connected servers.
Client (Claude Code, Cursor, VS Code, Windsurf...)
|
+-- mentu-mcp (12 meta-tools, ~1,000 tokens)
|
+-- mcp_discover search across all servers
+-- mcp_provision intent-based provisioning
+-- mcp_call forward tool calls
+-- mcp_execute V8 sandboxed code execution
+-- mcp_do autonomous orchestration (API)
+-- mcp_cortex diagnostic advisory (API)
+-- mcp_consult knowledge query (API)
+-- mcp_reload reload configuration
+-- mcp_elevate trust elevation
+-- mcp_register runtime server registration
+-- mcp_skill_* skill catalog (2 tools)
|
+-- child-server-1 (50 tools)
+-- child-server-2 (30 tools)
+-- child-server-N (...)Quick Start
# Install globally
npm install -g mentu-mcp
# Auto-configure for your MCP client
mentu-mcp --init
# Verify
mentu-mcp --version--init detects your MCP client (Claude Code, Cursor, VS Code, Windsurf, Zed, Claude Desktop, Gemini CLI, Amazon Q) and writes the appropriate configuration file.
Supported Clients
| Client | Config file | Auto-detected |
|--------|-------------|:---:|
| Claude Code | .mcp.json | Yes |
| Cursor | .cursor/mcp.json | Yes |
| VS Code | .vscode/mcp.json | Yes |
| Windsurf | ~/.windsurf/mcp.json | Yes |
| Zed | ~/.config/zed/settings.json | Yes |
| Claude Desktop | claude_desktop_config.json | Yes |
| Gemini CLI | .gemini/settings.json | Yes |
| Amazon Q | .amazonq/mcp.json | Yes |
| Any MCP-compatible client | Manual .mcp.json | -- |
Tools
| Tool | Description | Requires API |
|------|-------------|:---:|
| mcp_discover | Search across all servers' tool catalogs | -- |
| mcp_provision | Intent-based server provisioning from npm | -- |
| mcp_call | Forward a tool call to a child server | -- |
| mcp_execute | Run code in a V8 sandbox | -- |
| mcp_reload | Reload .mcp.json configuration at runtime | -- |
| mcp_elevate | Elevate trust level for auto-provisioned servers | -- |
| mcp_register | Register a new server at runtime | -- |
| mcp_skill_discover | Search the skill catalog | -- |
| mcp_skill_advise | Get skill recommendations for a task | -- |
| mcp_do | Autonomous multi-server orchestration | Yes |
| mcp_cortex | Diagnostic advisory | Yes |
| mcp_consult | Knowledge query | Yes |
Configuration
Add child MCP servers to ~/.mcp.json (or use --init to generate):
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["-y", "@playwright/mcp@latest"]
},
"my-server": {
"command": "node",
"args": ["./path/to/server.js"]
}
}
}mentu-mcp spawns child servers lazily on first use, manages connection pooling with circuit breakers, and tears them down on idle timeout.
API Intelligence
9 of 12 tools run entirely locally with no network calls. 3 tools (mcp_do, mcp_cortex, mcp_consult) are backed by the mentu intelligence API for perception scoring, judgment evaluation, and diagnostic advisory.
Without an API key: all local tools work normally. API-backed tools return a clear error explaining the requirement.
With an API key:
export MENTU_API_KEY=your_keyOr set it in your .mcp.json:
{
"mcpServers": {
"mentu": {
"command": "npx",
"args": ["-y", "mentu-mcp"],
"env": {
"MENTU_API_KEY": "your_key"
}
}
}
}100 free API calls per day. No credit card required.
Intent Routing
Map natural-language intents to specific tools in your config:
{
"intents": {
"crawl": "crawlio.start_crawl",
"analyze": "crawlio.analyze_page",
"browse": "playwright.execute"
}
}When mcp_provision receives a request matching an intent, it routes directly to the mapped tool without searching.
Environment Variables
| Variable | Purpose | Required |
|----------|---------|:---:|
| MENTU_API_KEY | API key for intelligence features | No |
Contributing
See CONTRIBUTING.md for development setup, project structure, and pull request guidelines.
Security
See SECURITY.md for reporting vulnerabilities.
