Watch: How Agent A explores and documents, then Agent B one-shots the same task - 2.6x faster!
📖 Read the story: Building CodeMesh - From Idea to Self-Improving Intelligence
CodeMesh lets AI agents write TypeScript code to orchestrate ANY MCP server. One prompt. One code block. Multiple servers working together.
Instead of exposing dozens of individual tools, CodeMesh provides just 3 tools:
discover-tools- See what's available (context-efficient overview)get-tool-apis- Get TypeScript APIs for specific toolsexecute-code- Execute TypeScript that calls multiple tools
When agents encounter unclear tool outputs, CodeMesh forces documentation before proceeding. This creates compound intelligence - each exploration helps ALL future agents.
Proven: Agent A explored and documented. Agent B one-shot the same task. 2.6x faster! 🚀
claude mcp add codemesh npx -y codemeshOr manually add to your Claude Desktop MCP settings:
{
"mcpServers": {
"codemesh": {
"command": "npx",
"args": ["-y", "codemesh"]
}
}
}Create a .codemesh/config.json file in your project directory to configure which MCP servers CodeMesh should connect to:
{
"logging": {
"enabled": true,
"level": "info",
"logDir": ".codemesh/logs"
},
"servers": [
{
"id": "filesystem",
"name": "File System",
"type": "stdio",
"command": ["npx", "@modelcontextprotocol/server-filesystem", "/path/to/directory"],
"timeout": 30000
},
{
"id": "brave-search",
"name": "Brave Search",
"type": "stdio",
"command": ["npx", "-y", "@modelcontextprotocol/server-brave-search"],
"env": {
"BRAVE_API_KEY": "${BRAVE_API_KEY}"
}
},
{
"id": "weather",
"name": "Weather Server",
"type": "http",
