Read API

Everything the site shows can be read as JSON. No account and no key; answers are cached for a few minutes.

GET /api/v1/search

Search and filter the catalogue: q, category, client, runs, package, licence, updated, sort, cursor, limit.

GET /api/v1/servers/{namespace}/{name}

One server as its page shows it.

GET /api/v1/servers/{namespace}/{name}/install

Its install configs and the install prompt (?client= for one client; Accept: text/markdown for the prompt alone).

GET /api/v1/categories

The categories with their counts.

GET /v0.1/servers

The Official MCP Registry's read API: each entry as the registry gives it, our facts under _meta["directory.plug/portal"]. cursor, limit, search, updated_since, version=latest.

GET /v0.1/servers/{name}/versions[/{version}]

One server’s versions, or one of them (latest for the newest).

POST /mcp

Our MCP server (streamable HTTP): search_servers, get_server, find_tool, install_config.

GET /data

The monthly data export: its files, sizes, checksums and licences.

GET /data/clients.json

The MCP clients we write configs for: each one’s config dialect, the install classes it runs (install_class in the export and in _meta), how far its config is checked, its docs, and what its page says about it (publisher, kind, platforms, price, licence, our description) with the source of each fact.

Errors are application/problem+json. The base address is https://plug.directory. Please keep to a few requests a second, as the acceptable use policy asks; for the whole catalogue, take the data export.