Code Executor
Orchestrates multiple MCP tools on-demand, drastically reducing token usage for AI agent contexts.
The Code Executor addresses the critical problem of token bloat encountered when AI agents, such as Claude Code, interact with a large number of MCP tools. By disabling direct access to numerous MCPs and funnelling all tool calls through two core executors (TypeScript and Python), it achieves up to 98% token savings. This system dynamically accesses other MCPs as needed, validating all parameters against live schemas using AJV to ensure correctness and provide detailed error feedback. It's ideal for complex AI setups managing a mix of local, remote, and API-based tools, offering robust security, auditing, and rate-limiting features.
主要功能
01TypeScript (Deno) and Python code execution
02Sandboxed execution, allowlisting, audit logs, and rate limiting for security
03Deep AJV-based parameter validation with disk-cached schemas
04High quality with extensive TypeScript, 139 tests, and 98%+ coverage
0556 GitHub stars
06Auto-discovery and environment variable configuration for MCP integration
使用案例
01Managing 3 or more MCP servers to avoid context bloat
02Integrating a mix of local, remote, and API-based tools for AI agents
03Requiring audit logs, allowlisting, and rate limiting for tool interactions