@daddyapi/mcp

Serve a daddyapi SiteSpec as an MCP server, so AI agents can read a bot-friendly website through typed, read-only tools.

Most people use it through the CLI:

npx -y @daddyapi/cli mcp ./quotes.yaml

From code:

import { loadSiteSpec } from '@daddyapi/spec';
import { createMcpServer, serveMcpStdio } from '@daddyapi/mcp';

const spec = loadSiteSpec('quotes.yaml');

await serveMcpStdio(spec);              // stdio, for Claude Desktop, Claude Code, Cursor, ...
const server = createMcpServer(spec);   // or connect it to any MCP transport yourself

Each resource in the spec becomes one tool:

• Inputs: the spec's {param} path parameters (required strings), pages (bounded by maxPagesLimit, default 20), limit, offset, and fields as an enum of the resource's field names.
• Output: { data, warnings, meta } as structured content with an output schema derived from the field transforms, plus the same JSON as text.
• Annotations: readOnlyHint, idempotentHint and openWorldHint are set; nothing a tool does changes the site.
• Errors: invalid arguments, robots.txt blocks and site failures come back as tool errors (isError: true) with a readable message.

The server also exposes the OpenAPI document as the MCP resource daddyapi://<spec name>/openapi.json.

Options: fetcher, hooks, maxPagesLimit, toolPrefix (namespace tool names when running several specs), version.