Ambit: see what you and your AI agents can do, and what one more step would open

What you, your agents and your machines can jointly do is your ambit. It is spread across LLM providers, MCP servers, local CLI tools, skills, credentials and more than one machine. Ambit reads the configs of Claude Code, Cursor, OpenCode, Windsurf, Gemini CLI, Claude Desktop, Codex CLI, Cline, Roo Code, Continue, Zed and VS Code, and draws that reach as one map, so you can see how to widen it.

What it answers

• What is worth setting up next, ranked by what it would open and the attention it would save.
• Which pieces pass their checks, and what is one dependency away.
• What compound abilities emerge when two tools are combined.
• What is configured but failing, and what breaks downstream if a model, MCP server, tool or credential goes away.

You ask from the terminal. Your agents ask over MCP: Ambit is itself an MCP server, with a ten-tool profile for an agent's context and sixty tools in all. An agent can propose a configuration change; applying one takes an approval, made outside MCP. Everything stays in a local SQLite file.

To run it on your own configs: npx ambit-cli, with Node 22.18 or newer on macOS or Linux. The interactive map needs JavaScript. Read about it instead:

• Live demo, the map on example data
• Guide: install, the CLI, and connecting it to Claude Code or OpenCode
• FAQ: what it reads, and what leaves your machine
• Where each AI agent keeps its MCP config: Claude Code, Cursor, Codex CLI, Goose, Kiro and more
• What breaks if an MCP server goes down: finding single points of failure
• Auditing an agent's MCP servers: declared, working, permitted, and used
• Reference: the model and every MCP tool
• Ambit and Jev: typed judgment on the capability map
• All documentation
• Source on GitHub, MIT licensed