Caselaw MCP Server

A local MCP server providing real-time access to UK court judgments from Find Case Law at The National Archives.

Overview

Property	Value
Source	caselaw.nationalarchives.gov.uk (Open Justice Licence)
API key	Not required for search/lookup; summarise_judgment needs a user-supplied Anthropic API key
Transport	stdio (runs as a local process)
Location	mcp-servers/caselaw/src/index.ts
User-Agent	ai-legal-uk-caselaw-mcp/1.0
Cache	Tiered in-memory, 500-entry LRU
Fetch timeout	30 seconds

Open Justice Licence

Find Case Law is published by The National Archives under the Open Justice Licence. Judgment text may be used for the purposes described in that licence; it must not be used to identify or contact parties, and computational analysis is permitted on an open basis. Outputs from these tools are AI-assisted research, not legal advice — England & Wales only.

Running the Server

cd mcp-servers/caselaw
npm install
npm run dev     # start the server (tsx src/index.ts)

For production builds:

npm run build   # compiles to dist/
npm test        # build + run tests

Automatic Startup

When configured in .mcp.json, Claude Code starts the server automatically when a skill needs it. You do not need to start it manually.

Tools

The server exposes 9 tools. Three search the Atom feed, two read a specific judgment, one finds judgments by statute section, one downloads the PDF, one extracts body text, and one produces an AI summary.

#	Tool	Purpose	Key parameters
1	search_caselaw	Free-text search across judgments	query (req), court, limit
2	get_recent_judgments	Newest judgments, most recent first	court, limit
3	search_by_party	Judgments involving a named party	party_name (req), court, limit
4	search_by_judge	Judgments decided by a named judge	judge_name (req), court, limit
5	lookup_judgment	Full metadata + body text for one judgment	uri (req)
6	get_judgments_for_section	Judgments citing a statute section	legislation_id (req), section_number (req)
7	download_judgment_pdf	Save a judgment PDF to disk	uri (req), output_path
8	extract_judgment_text	Body text as structured Markdown	uri (req)
9	summarise_judgment	AI summary of a judgment	uri (req), style, apiKey (req)

1. search_caselaw

Search Find Case Law for judgments by free-text query.

Parameter	Type	Required	Description
query	string	Yes	Free-text search query
court	string	No	Court code to filter by (e.g. ewca, uksc, ewhc/pat)
limit	number	No	Max results, 1–50 (default 20)

Returns: A Markdown list of matching judgments, each with name, neutral citation (where detectable), last-updated date, and URI.

Example: search_caselaw("frustration of contract", court="ewca")

2. get_recent_judgments

Get the most recently published judgments, optionally filtered by court (ordered newest-first via order=-date).

Parameter	Type	Required	Description
court	string	No	Court code
limit	number	No	Max results, 1–50 (default 20)

Returns: A Markdown list of the latest judgments, with the same fields as search_caselaw.

3. search_by_party

Search for judgments involving a named party (claimant or defendant).

Parameter	Type	Required	Description
party_name	string	Yes	Name of the party
court	string	No	Court code
limit	number	No	Max results, 1–50 (default 20)

Example: search_by_party("Uber BV", court="uksc")

4. search_by_judge

Search for judgments decided by a named judge.

Parameter	Type	Required	Description
judge_name	string	Yes	Name of the judge
court	string	No	Court code
limit	number	No	Max results, 1–50 (default 20)

Example: search_by_judge("Lord Leggatt")

5. lookup_judgment

Get full metadata and body text for a specific judgment by its URI.

Parameter	Type	Required	Description
uri	string	Yes	Judgment URI, e.g. uksc/2024/42 or tna.t7sh6v3m

Returns: A Markdown record with title, neutral citation, court, date, judges, parties, URI, and the judgment body. The URI is parsed from the canonical Akoma Ntoso XML at …/data.xml. If only a PDF exists, the body falls back to a note recommending extract_judgment_text.

Example: lookup_judgment(uri="uksc/2024/42")

6. get_judgments_for_section

Find judgments that cite a specific statute section. This is the bridge between the legislation and caselaw servers — give it a legislation ID and section number and it returns judgments that have considered that provision.

Parameter	Type	Required	Description
legislation_id	string	Yes	Legislation ID, e.g. ukpga/1996/18
section_number	string	Yes	Section number, e.g. 94 or 10A

How it works: the tool first fetches the Act's title from legislation.gov.uk, then runs a multi-strategy search — full title (e.g. "Employment Rights Act 1996" section 94) plus a derived short name (e.g. "ERA 1996" s.94) — and de-duplicates the combined results by URI.

Example: get_judgments_for_section(legislation_id="ukpga/1996/18", section_number="94")
         → Judgments citing Employment Rights Act 1996, s.94 (unfair dismissal)

7. download_judgment_pdf

Download a judgment PDF to disk and return the local file path.

Parameter	Type	Required	Description
uri	string	Yes	Judgment URI, e.g. uksc/2024/42
output_path	string	No	Target file path. Defaults to ./{uri-slug}.pdf in the current working directory

Returns: A confirmation of the bytes written and the absolute path. The URI slug is sanitised (path traversal stripped) before writing.

8. extract_judgment_text

Extract the body text of a judgment as structured Markdown. Useful when the canonical XML body is thin or the judgment is PDF-only.

Parameter	Type	Required	Description
uri	string	Yes	Judgment URI

Strategy (in order):

1. XML — fastest; reads the Akoma Ntoso data.xml body.
2. HTML → MarkItDown — fetches the judgment HTML and converts it to Markdown with headings and paragraphs preserved.
3. PDF → MarkItDown — downloads data.pdf and converts it to Markdown.

MarkItDown dependency

Strategies 2 and 3 shell out to Python's markitdown. The server tries a project .venv/bin/python3 first, then system python3. If MarkItDown is unavailable and the XML body is empty, the tool reports that the judgment could not be extracted in a machine-readable format.

9. summarise_judgment

Generate an AI summary of a judgment. This is the only tool that calls a model and the only one that requires an API key.

Parameter	Type	Required	Description
uri	string	Yes	Judgment URI
style	enum	No	brief (default), detailed, or plain-english
apiKey	string	Yes	User-supplied Anthropic API key

API key required

summarise_judgment instantiates the Anthropic SDK with your apiKey and sends the judgment body to a Claude model (claude-sonnet-4-6) for summarisation. The other eight tools hit only the public Find Case Law endpoints and need no key. The key is used in-process for the request and to namespace the cache; it is not persisted to disk.

Style	What you get
brief	3–4 sentences: the holding and a one-line reason
detailed	300–500 words: holding, ratio decidendi, key facts, procedural history, with paragraph citations
plain-english	A non-lawyer explanation of what happened, what was decided, and why it matters day-to-day

If the judgment has no XML body text, the tool reports that it cannot summarise without first extracting PDF text — run extract_judgment_text (or download_judgment_pdf) first.

Example: summarise_judgment(uri="uksc/2024/42", style="plain-english", apiKey="sk-ant-…")

Court Codes

Court codes filter searches to a specific court or tribunal. Common values:

Code	Court / Tribunal
uksc	UK Supreme Court
ewca	Court of Appeal (England & Wales)
ewhc	High Court (England & Wales)
ewhc/pat	High Court — Patents Court
ukeat	Employment Appeal Tribunal
ukut	Upper Tribunal
ukftt	First-tier Tribunal

Caching

The server keeps a tiered in-memory cache (max 500 entries, LRU eviction) so repeated lookups during a session do not re-hit Find Case Law.

Data	TTL
Search results (Atom feed)	1 hour
Judgment lookups (data.xml)	7 days
AI summaries	30 days

Cache Behaviour

Cache hits are logged to stderr as [caselaw] Cache hit: <url>. AI summaries are cached per (uri, style, apiKey-hash), so a repeated summary request with the same key returns instantly without another model call. The cache is per server process and resets on restart.

Error Handling

Scenario	Behaviour
Judgment not found	Returns Error: Not found: <url>
Rate limited	Returns Error: Rate limited (429) — try again in 5 minutes
Network timeout	Returns Error: Timeout fetching <url>
HTTP error	Returns Error: HTTP <status> fetching <url>
PDF unavailable	download_judgment_pdf returns Error: PDF not available for <uri>
No machine-readable text	extract_judgment_text reports that XML, HTML, and PDF strategies all failed

Related

• /legal caselaw — the CLI command that drives these tools to research and summarise judgments.
• Lex Remote MCP Server — a complementary remote server with 63,000+ judgments and semantic search; use it for AI-ranked case-law discovery alongside this local server's direct lookups.
• UK Legislation MCP Server — pairs with get_judgments_for_section to move from statute to the cases that have applied it. See also /reference/legislation.