Agents
Agent Overview
Clean works with Claude Code, Cursor, Claude Desktop, Antigravity, and Codex via MCP — locally over stdio or through the cloud.
Clean connects to AI coding agents through the Model Context Protocol (MCP), an open standard for giving AI tools structured access to external services.
Once connected, your agent gains Clean's full set of MCP tools — search_code, expand_result, get_source, get_file_tree, list_repos, and (on the local edition) index_repo and delete_repo.
Two transports
How your agent connects depends on which edition you use:
| Edition | Transport | How it connects |
|---|---|---|
| Local (open source) | stdio | The agent launches python -m clean.local.mcp_server and talks to it over stdin/stdout. No URL, no key. |
| Cloud | SSE / HTTP | The agent connects to https://api.tryclean.ai/mcp with an API key and org slug. |
Pick a tab on each agent's page to see the matching config.
Supported agents
| Agent | Local config location | Cloud config location |
|---|---|---|
| Claude Code | .mcp.json (or claude mcp add) | ~/.claude.json |
| Cursor | ~/.cursor/mcp.json | ~/.cursor/mcp.json |
| Claude Desktop | Claude Desktop config (command) | Claude Desktop config (stdio proxy) |
| Antigravity | ~/.gemini/antigravity/mcp_config.json | ~/.gemini/antigravity/mcp_config.json |
| Codex | ~/.codex/config.toml (or codex mcp add) | ~/.codex/config.toml |
Any MCP-compatible client works — these just have ready-made snippets.
Getting your config
- Local: copy the stdio snippet from each agent's page, or see Running the server.
- Cloud: the dashboard generates the correct config block for each agent. Go to mcp.tryclean.ai/dashboard/onboarding and open the MCP Config step, or visit Dashboard → API Keys to get a config with a specific key.