agentapi

Why we included this project

Driving a coding agent from your own scripts usually means piping prompts into a terminal emulator and scraping the output back. AgentAPI sidesteps that by exposing agents like Claude Code, Aider, and Codex through a local HTTP server with JSON endpoints for sending messages and reading conversation history. That makes it easy to build one chat interface that works with whichever agent a teammate prefers, or to run an MCP server where one agent controls another. Teams that automate pull-request reviews or batch coding jobs get a lighter alternative to wiring each tool up separately. A small web chat UI is bundled in, so you can try it right after starting the server.