Build against
the archive.
Gold Radar speaks MCP. Point Claude, Cursor, ChatGPT or a custom agent at one endpoint and let it file work as you — with the prompt, model and credit intact.
https://www.goldradar.ai/mcpStateless Streamable HTTP. POST JSON-RPC with an Authorization: Bearer gr_… header.
Tokens are minted per maker in the Studio. They are stored as hashes, scoped to one creator, and revocable at any time.
Mint a tokenClient config
{
"mcpServers": {
"goldradar": {
"url": "https://www.goldradar.ai/mcp",
"headers": {
"Authorization": "Bearer gr_your_token"
}
}
}
}The exact config shape varies by client; the URL and header are the two constants. Tokens can also be sent from server-side MCP clients without a browser.
Tools
whoamiVerify the token and its maker file.
publish_artifactFile a public URL with its prompt, model, tools and tags.
list_my_artifactsList the works filed by this maker.
get_artifactRead one work by slug.
update_artifactEdit the work's metadata or status.
delete_artifactRemove the work and its uploaded media.
Machine-readable discovery
/.well-known/ai-catalog.jsonapplication/ai-catalog+jsonMCP Server Card/mcp/server-cardapplication/mcp-server-card+jsonllms.txt/llms.txttext/plainMCP Registryhttps://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.zocoi/goldradario.github.zocoi/goldradarAgents that follow the AI Catalog convention should read the catalog first; it points at the Server Card, which describes the endpoint, transport and required header.