Quickstart
From zero to semantic code search in 5 minutes.
Overview
Here's the full pipeline:
- Sign up at tryclean.ai
- Grab your API key from the dashboard
- Run
npx @tryclean/create— the CLI sets everything up - Enter your API key when prompted
- Select which coding agents to connect (Claude, Cursor, Antigravity, Codex)
- (Cursor only) Paste the generated MCP config into Cursor's settings
- Start chatting — your agent can now search your codebase semantically
Step 1 — Sign up
Go to tryclean.ai and create an account.
Step 2 — Get your API key
After signing in, open the dashboard and copy your API key. It looks like clean_sk_....
Step 3 — Run the installer
The interactive CLI will guide you through setup — provisioning your Clean instance, generating config files, and writing MCP configs to your agent's settings directory.
Step 4 — Enter your API key
When the CLI asks for your API key, paste the clean_sk_... key you copied from the dashboard.
Step 5 — Select your agents
The CLI will detect which agents are installed and let you pick which ones to connect:
- Claude Code — config written automatically to
~/.claude.json - Cursor — config written to
~/.cursor/mcp.json, but you need to paste it into Cursor's MCP settings (see Cursor setup) - Antigravity — config written to
~/.gemini/antigravity/mcp_config.json - Codex — config written to
~/.codex/config.toml
Step 6 — (Cursor only) Paste the MCP config
Cursor requires you to manually add the MCP server in its settings UI. See Cursor setup for the exact steps.
Step 7 — Try it
Ask your agent:
"Search for the authentication logic"
The agent will call search_code through MCP and return semantically matched functions with their full call graph context.