whatdotheyknow-mcp

A Model Context Protocol server for UK Freedom of Information research. Connects AI assistants to WhatDoTheyKnow — the UK's largest FOI request platform — to search requests, read responses, look up public authorities, and draft new requests.

Fork note: This fork adds six new tools that expose Resources and scrape HTML correspondence:

• get_request_detail, get_user_requests, get_authority_detail expose existing MCP Resources as callable tools (essential for Claude.ai connectors)
• get_request_v2, get_request_messages attempt v2 API endpoints (may not be available on all Alaveteli instances)
• get_request_correspondence scrapes and parses actual message bodies from HTML pages using browser impersonation

This is essential for MCP clients that can only invoke tools, not access Resources directly. Upstream: paulieb89/whatdotheyknow-mcp

Tools

Tool	Description
search_request_events	Full-text search of FOI requests and responses via WhatDoTheyKnow's Atom feed. Supports structured expressions (status:successful, body:"Liverpool City Council").
search_authorities	Search UK public authorities by name. Returns slug for use with other tools.
get_request_feed_items	Fetch the event timeline (sent, response, clarification) for a specific FOI request.
get_request_detail	NEW — Fetch the full JSON detail for an FOI request, including all correspondence text, status, dates, and attachments.
get_request_v2	NEW — Attempt to fetch full request via Alaveteli v2 API (may not be available on all instances).
get_request_messages	NEW — Attempt to fetch dedicated message list via v2 API (may not be available on all instances).
get_request_correspondence	NEW — Scrape and parse actual message bodies from HTML request pages using browser impersonation + BeautifulSoup. This provides the actual text that the JSON API omits.
get_user_requests	NEW — Fetch a user's profile and list of all their FOI requests with titles, slugs, statuses, and authorities.
get_authority_detail	NEW — Fetch full detail for a public authority, including contact info, description, and recent requests.
build_request_url	Build a prefilled WhatDoTheyKnow request URL for a given authority and topic.
create_request_record	Create a request via the write API (requires WDTK_API_KEY).
update_request_state	Update user-assessed state of a request (requires WDTK_API_KEY).

Resources

URI template	Returns
wdtk://authorities/{authority_slug}	Authority profile JSON
wdtk://requests/{request_slug}	FOI request detail JSON
wdtk://users/{user_slug}	User profile JSON
wdtk://requests/{request_slug}/feed	Request event Atom feed
wdtk://users/{user_slug}/feed	User activity Atom feed
wdtk://authorities/all.csv	Full CSV of all UK public authorities

Prompts

Prompt	Description
draft_foi_request	Draft a narrow, specific FOI request for a given authority and topic.

What's new in this fork

The upstream MCP server defines Resources for reading request detail, user profiles, and authority info — but many MCP clients (including Claude.ai's connector system) can only call tools, not access Resources. This fork adds three tools that wrap those Resources:

get_request_detail(request_slug)

Returns the complete JSON for an FOI request, including:

• Original request text and all response correspondence
• Current status and described state
• Dates (created, updated)
• Public body info
• Full info_request_events array with every message

Example: get_request_detail("payroll_system_synchronization_f_2")

get_user_requests(user_slug)

Returns a user's profile and all their FOI requests, including:

• Display name and about text
• Array of info_requests with titles, slugs, statuses, and authorities

Example: get_user_requests("g_spinks")

get_authority_detail(authority_slug)

Returns full authority profile, including:

• Name, description, home page
• Request email and tag string
• Recent requests made to this authority

Example: get_authority_detail("essex_police")

get_request_v2(request_id) & get_request_messages(request_id)

Attempt to fetch request data via the Alaveteli v2 API. These endpoints may not be available on all Alaveteli instances (including WhatDoTheyKnow). The request_id is the numeric ID found in v1 API responses.

Caveat: WhatDoTheyKnow currently returns errors on these endpoints (they may be admin-only or not implemented on this instance).

get_request_correspondence(request_slug) ⭐

Scrapes and parses the actual message bodies from the HTML request page. This solves the original problem — the JSON API truncates or omits message text, but this tool fetches the HTML page using browser impersonation (curl-cffi with Safari 17 user agent) and parses it with BeautifulSoup to extract the complete correspondence.

Returns:

• title: request title
• status: current status
• messages: list of message objects with type, date, from, and body
• url: link to the original request
• note: clarification that this is HTML-scraped data

Example: get_request_correspondence("payroll_system_synchronization_f_2")

Note: This tool performs real HTTP requests and HTML parsing, so it's slightly slower than JSON API calls. It respects rate limits and uses browser impersonation to avoid blocks.

Connect

Hosted (no install)

{
  "mcpServers": {
    "whatdotheyknow": {
      "type": "http",
      "url": "https://whatdotheyknow-mcp.fly.dev/mcp"
    }
  }
}

Local (uvx)

{
  "mcpServers": {
    "whatdotheyknow": {
      "type": "stdio",
      "command": "uvx",
      "args": ["whatdotheyknow-mcp"]
    }
  }
}

Environment variables

Variable	Required	Description
WDTK_API_KEY	Optional	Enables create_request_record and update_request_state write tools

Upstream API and Licence

Source	API	Licence	Auth
WhatDoTheyKnow	www.whatdotheyknow.com	OGL v3	None (read) / API key (write)

Data is sourced directly from the WhatDoTheyKnow public API. The platform is operated by mySociety.