Which artifact does your tool use?
Every project ships four artifacts from its Connect page. Different tools consume different ones — picking right is the difference between "wired up in 60 seconds" and "fighting context all afternoon."
| Tool | Primary artifact | Why |
|---|---|---|
| Cursor | mcp.json | Native MCP support; tools appear in the agent toolbar. |
| Claude Code | .mcp.json | Native MCP support; agent can call your platform tools. |
| Windsurf | mcp_config.json | Cascade calls the platform tools mid-flow. |
| v0.dev | llms.txt | No MCP yet — paste the brief as up-front context. |
| Lovable | llms.txt | Pin the brief as a knowledge document. |
| Bolt | llms.txt + SDK | Drop the brief into the first prompt; the SDK ships with the generated project. |
Cursor
Per-workspace .cursor/mcp.json that registers every endpoint of your project as a tool the agent can call directly from chat or Composer.
Open guide →Claude Code
Drop .mcp.json at the repo root, export the API key, and Claude gets first-class access to your schema and endpoints from the CLI.
Open guide →Windsurf
Paste the MCP config in Windsurf settings, and Cascade can change your schema and write the matching client code in a single flow.
Open guide →v0.dev
Paste llms.txt as the leading context block so v0 generates components that fetch from real endpoints with the right shape and auth header.
Open guide →Lovable
Pin llms.txt as a knowledge document so every iteration sees the real backend, and pair it with a small init snippet for the chisel.to client.
Open guide →Bolt
Front-load llms.txt and ask Bolt to scaffold a full-stack project that calls your chisel.to backend, with auth, data and realtime wired from the start.
Open guide →mcp.json. For prompt-only tools (Replit AI, ChatGPT Canvas, etc.), the llms.txt brief works the same way it does for v0 and Lovable.