Descrybe Legal Engine

Name: Descrybe Legal Engine MCP Server
Availability: InStock
Author: Descrybe Legal Engine

Productivity13 tools

Search U.S. case law, statutes and regulations, resolve citations, verify quotes and check later treatment from your AI agent. 13 tools, OAuth sign-in, paid plan required, legal research support — not legal advice.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on 16 August 2026.

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect Descrybe Legal Engine via MCP

https://mcp.descrybe.com/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Descrybe Legal Engine Tools & Capabilities (13)

analyze_legal_questionRead-only

Assess a broad or fact-dependent legal research question before searching to identify the jurisdiction, facts, legal issue, and procedural context needed for useful research. Returns ready_for_research or needs_clarification with up to three short clarifying questions.

search_cases_by_conceptRead-only

Use this first when the user wants to find leading or relevant cases about a legal issue, doctrine, fact pattern, jurisdiction, claim, defense, or topic and does not already have a Descrybe case_id. Results are authority-ranked by default to surface stronger legal authorities before ordinary text matches. Use date sorting only when the user asks for recent or newest cases. Each result includes case_id, the Descrybe canonical identifier accepted by known-case tools.

search_laws_and_rulesRead-only

Semantic/concept search over U.S. non-case primary law, including statutes, regulations, and constitutional provisions. Useful for legal rules outside judicial opinions when the issue, rule, citation, or topic can be described in natural language. Results are ranked by concept similarity and returned with compact matched source passages.

search_case_textRead-only

Searches the full text of judicial opinions by exact words or phrases across Descrybe's case-law text index. Useful when specific language matters, such as terms of art, quotes, statutory phrases, or uncommon factual wording. Each result includes case_id, the Descrybe canonical identifier accepted by known-case tools.

find_case_from_referenceRead-only

Resolves a citation, case name, case caption, party name, party-name variant, docket number, short cite, quote, or mixed reference into a Descrybe case or candidate set. Use this for party-name or docket-number case lookup when the user does not already have a Descrybe case_id. Resolved cases and candidates include case_id, the Descrybe canonical identifier accepted by known-case tools. The context_text field can improve short, bare, or ambiguous references.

extract_case_referencesRead-only

Extracts case-law citations and references from legal text provided in the text argument. Returns one item per unique resolved case or best unresolved reference cluster, with occurrence counts, raw citations seen, and tagged excerpts. When resolve=true, results include compact case-identity resolution and a quick treatment indicator for each resolved case, including case_id when a Descrybe case is resolved. This does not verify quotes, analyze support, or return a full citing-case treatment map.

get_case_summaryRead-only

Retrieves the precomputed standard or simplified summary for a known Descrybe case's primary opinion. Requires a Descrybe case_id; for legal research, prefer the standard summary unless the user asks for a simplified or plain-language explanation.

get_case_passagesRead-only

Returns a small ranked list of focused raw passages from a known Descrybe case's primary opinion. Requires a Descrybe case_id and a focus issue, phrase, rule, quote, or fact pattern. This is not a full-opinion fetch and does not return every occurrence.

get_case_detailsRead-only

Retrieves case metadata, summary availability, opinion inventory, and treatment summary for a known Descrybe case. Requires a Descrybe case_id.

check_case_statusRead-only

Checks a known Descrybe case for a quick treatment/status signal before relying on it. Requires a Descrybe case_id. Returns whether later cases appear to treat the case positively, negatively, cautiously, or neutrally, plus a rough authority-weight label. This is a fast status check, not a full citing-case map and not a forum-specific binding analysis.

find_cases_that_citeRead-only

Finds later cases that cite a known Descrybe case. Requires a Descrybe case_id. Useful for updating research, finding treatment, or locating newer authority that follows, distinguishes, or discusses the case. Returned citing cases include case_id when a Descrybe case is identified.

verify_quoteRead-only

Verifies whether quoted language appears in a known Descrybe case and locates the matching opinion. Requires a Descrybe case_id and exact or near-exact quote text. Returns whether the quote was found, where it matched, and which opinion contained it.

get_case_pdfActs

Creates or reuses a generated Descrybe case PDF asset and returns its public URL. Requires a Descrybe case_id. Returns URL/job metadata or preparation status, not PDF bytes; present the URL to the user instead of trying to download the file.

Read from the server on 2026-08-16, including each tool's own safety annotations.

Limits

Descrybe does not validate its own AI-generated output. Its terms state it uses third-party AI models to summarize and make searchable U.S. primary law, and that it *"does not review or validate the Outputs"* for accuracy, truthfulness, reliability or freedom from error. The same section says outputs may be inaccurate, misleading or erroneous, and that use is at your own risk. Descrybe's FAQ says output should support rather than replace your own research and professional judgment, and to always verify against the original opinion.
A summary is not the authority. get_case_summary returns a precomputed summary; its simplified option is explicitly for plain-language explanation, and the tool tells assistants to prefer the standard summary for research fidelity. Cite the opinion, not the précis.
Some state law is deliberately excluded, including in large jurisdictions. Descrybe's coverage page marks statutes restricted for Arkansas, Georgia, Mississippi and Tennessee, and regulations restricted for California, New Jersey, New York and Vermont — and states restricted sources are excluded from the corpus because their access terms limit collection, storage, reuse or commercial use. Vermont court rules are restricted too; Arizona, Colorado, Maryland and Wyoming court rules are partial; and there are no federal session laws. Maryland regulations run on a delay — Descrybe waits for the public-domain COMAR version released 180 days after publication. This is the gap most likely to catch a researcher out, because case-law coverage is broad enough to make the corpus feel complete.
Currency varies by court. Descrybe's FAQ states coverage varies by court, with some publishing immediately and others delaying. Its coverage page publishes a first and most recent opinion date per court. A very recent decision may not be in yet, and neither the connector nor this page can tell you which courts are behind today — check the coverage page for the specific court.
U.S. primary law only. Every jurisdiction value across the schemas is a U.S. state, a federal case bucket or all. No foreign law, and no secondary sources such as treatises or practice guides.
The citator is a signal, not a conclusion. check_case_status describes itself as a fast status check, *"not a full citing-case map"* and not a forum-specific binding analysis. Descrybe's FAQ recommends confirming case status and reading the original opinion before relying on any authority.
Citation extraction stops short of verification. extract_case_references states it does not verify quotes, analyse support, or return a full citing-case treatment map. Use verify_quote separately for the quote check.
Passage retrieval is not full text. get_case_passages returns a small ranked list and its description says it is not a full-opinion fetch and does not return every occurrence.
It needs a paid plan. Standalone Legal Engine access is 25 dollars a month per Descrybe's pricing page, separate from your Claude, ChatGPT or Perplexity subscription, and each user needs their own — Descrybe's setup FAQ says access is required per person.
One tool is not annotated read-only. get_case_pdf creates or reuses a PDF asset server-side. It is not marked destructive and does not touch your data, but it is the one call that is not a pure read.
We did not exercise any tool. Our check was a read-only initialize and tools/list handshake. Everything above about behaviour comes from the server's own tool descriptions and Descrybe's published documentation, not from calling anything. In particular, we did not test result quality, coverage claims, or what an unauthenticated tool call returns.

Frequently asked questions

Does Descrybe Legal Engine give legal advice?

No. Descrybe's terms of service carry a section headed that Descrybe does not provide legal advice, stating outputs are for general information only, that Descrybe is not a law firm or lawyer, and that using it creates no attorney-client relationship. The server repeats this in its own connection instructions, telling assistants to treat results as legal research support rather than advice.

Do you need a paid Descrybe plan to use the MCP server?

Yes. Descrybe's Claude setup guide tells you to have active Legal Engine or Platform access before connecting, and lists being sent to checkout as the symptom of an account without it. Standalone Legal Engine access is 25 dollars a month on the pricing page. Anthropic's directory records the server as auth_required rather than authless.

What U.S. courts and dates does Descrybe cover?

Descrybe's coverage page reports 492 courts and tribunals holding 10,969,112 opinions, as of July 18, 2026. It splits into federal courts, state courts, bankruptcy courts, specialized and agency tribunals, and tribal courts. Per-court date ranges are published individually, and Descrybe states they vary with each court's publication practices and source availability.

Are Descrybe's case summaries written by AI?

Yes. Descrybe's terms state it uses third-party AI models to summarize U.S. primary law, and that it does not review or validate those outputs for accuracy. The same section warns outputs may be inaccurate or misleading. Descrybe's own guidance is to inspect the underlying opinion before relying on any summary rather than citing the summary itself.

Why does Claude ask for write permission on the Descrybe PDF tool?

Because get_case_pdf is the one tool the server does not annotate as read-only. Descrybe's Claude setup guide explains that the source PDF tool may be labelled differently because generating or displaying a PDF can technically count as creating a file. Its description says it creates or reuses a PDF asset and returns a URL, not file bytes.

Are any state statutes or regulations missing from Descrybe?

Yes. Descrybe's coverage page marks statutes restricted for Arkansas, Georgia, Mississippi and Tennessee, and regulations restricted for California, New Jersey, New York and Vermont. Restricted sources are excluded from the corpus because their access terms limit collection or commercial reuse. Maryland regulations are delayed: Descrybe waits for the public-domain COMAR release published 180 days later.

Do you need to enter a server URL to connect Descrybe?

No. Descrybe's setup FAQ says to install Descrybe from Claude's connector directory and that Claude handles the connection details. You open the listing, click Connect, sign in to Descrybe, then approve the tools. Descrybe recommends allowing all of them so the assistant can search, cite, check authority and verify quotes when needed.