Maverick MCP quickstart

Put real legal research inside ChatGPT Work and Claude Cowork.

Use one public, read-only remote MCP connection in ChatGPT Work, Claude Cowork, Codex, and Claude Code. Maverick searches statutes, regulations, and Maverick’s case-law index, then retrieves the complete primary text your agent needs to ground legal analysis.

One URL. No API key to paste. Name the connection Maverick Legal Research, choose OAuth, and use https://api.mavericklegalresearch.com/mcp.

Two-minute setup

Pick your product. Paste one URL.

An active Maverick trial or subscription is required. Interactive clients use browser OAuth; leave optional client IDs, client secrets, and API-key fields blank.

Copy this remote MCP URL

https://api.mavericklegalresearch.com/mcp

ChatGPT Work

As a workspace owner/admin, open ChatGPT app settings.
Choose Apps → Create and enable developer mode if prompted.
Name the app Maverick Legal Research, paste the URL above, and choose OAuth.
Click Scan Tools, complete Maverick sign-in, then create and publish the app.
Start a new Work chat and choose Maverick from Apps before sending your prompt.

Business creation/publishing is owner-only. Enterprise and Edu can grant developer access with RBAC, but an owner/admin still publishes the app. ChatGPT keeps an approved snapshot of tool definitions; refresh and republish after breaking schema changes.

OpenAI’s current custom MCP app guide

Claude Cowork

Open Claude connectors.
Choose Add → Add custom connector, then paste the URL above.
Click Connect, complete Maverick sign-in, and click Allow.
Start Cowork, click + → Connectors, and turn on Maverick for the task.

Team and Enterprise owners first add a Web connector under Organization settings → Connectors → Add → Custom → Web. Each member then connects their own Maverick seat. Cowork reaches remote connectors from Anthropic’s cloud, so use the public HTTPS URL—not localhost or a VPN-only host.

Anthropic’s current remote connector guide

Magic-link step: open only the newest Maverick email. It lands on Finish signing in; click Sign in to Maverick, then approve the client. That deliberate click prevents email-security scanners from using the one-time link first.

Codex desktop or CLI

codex mcp add maverick --url https://api.mavericklegalresearch.com/mcp

If it shows Not logged in, run codex mcp login maverick. Complete OAuth, then start a new task.

Claude Code

claude mcp add --transport http maverick https://api.mavericklegalresearch.com/mcp

Run /mcp, choose Maverick, complete OAuth, then start a new session.

Test the connection with an ordinary legal question

Paste this into a fresh Work or Cowork task with Maverick enabled:

Using Maverick, explain the initial-disclosure duty under Federal Rule of
Civil Procedure 26(a)(1) and relevance under Federal Rule of Evidence 401.

Success means the agent searches from that question, identifies the court-rule source family, and retrieves exact rule text without asking you for internal routing inputs.

The contract

Five focused tools. One simple research call.

Ask an ordinary-language question with an optional jurisdiction. Maverick handles source selection, normalization, Maverick ranking, exact resolution, and citator presentation.

Search legal authorities

search_legal_authorities({
  "query": "What counts as out-of-pocket damages under West Virginia consumer-protection law?",
  "jurisdiction": "WV"
})

The response includes statutes, cases, source scope, a factual retrieval assessment, and a Maverick citator object on every case. Jurisdiction accepts common names and aliases such as WV, West Virginia, and wva.

The five-tool catalog

search_legal_authorities — ordinary-language search
get_research_coverage — optional live-coverage discovery
get_statute_section — exact statutes, regulations, constitutions, and court rules
get_case_opinion — primary opinion text and Maverick citator status
lookup_citations — resolve citations in a block of text

Agent workflow

Search, read, then synthesize.

Search finds the best available authorities. The exact readers return the provisions and opinions an agent uses to explain the law, while retrieval and citator statuses state what Maverick found.

one jurisdiction per call

Keep comparisons clean.

For a state-law comparison, run the same neutrally framed issue once for each jurisdiction. Wait for both independently scoped result sets before comparing their governing statutes and cases.

factual status

Know what the retrieval found.

Every search is labeled responsive_authority_found, nearest_useful_leads, or no_results_found. Every returned case includes Maverick citator status and its as_of date.

Completion checklist

Each material issue has a focused authority search.
Every relied-on statute, regulation, and opinion has been fetched in full.
The jurisdiction, authority hierarchy, provenance, and retrieval assessment are accounted for.
The Maverick citator status of every relied-on case is reported.

Two-call comparison example

search_legal_authorities({
  "query": "Under this jurisdiction's consumer-protection law, what economic losses qualify as actual damages?",
  "jurisdiction": "WV"
})

search_legal_authorities({
  "query": "Under this jurisdiction's consumer-protection law, what economic losses qualify as actual damages?",
  "jurisdiction": "Ohio"
})

Compare only after both calls return.

Direct API

Calling the REST API with a key.

MCP is the fastest way in, but the HTTP API is fully supported for custom agents and server-side integrations. Every request needs two headers, and the second one is the one people miss.

1. Your API key — the secret

Sent as a bearer token. It starts with mvk_ and is shown exactly once, when you generate it in the portal. Treat it like a password: it identifies your firm and authorizes billing.

Authorization: Bearer mvk_...

2. `X-Maverick-User-Id` — not the key

A stable id you assign to each person using your key — any consistent string, one per seat. It is not secret and it is not your API key. It is how per-seat usage is attributed, and it is required: omit it and the request fails with 400 user_id_required, even with a valid key.

X-Maverick-User-Id: attorney-jdoe

A complete request

curl -X POST https://api.mavericklegalresearch.com/v1/search \
  -H "Authorization: Bearer mvk_..." \
  -H "X-Maverick-User-Id: attorney-jdoe" \
  -H "Content-Type: application/json" \
  -d '{"q": "premises liability open and obvious", "k": 10}'

Base URL https://api.mavericklegalresearch.com. Search is POST; document fetches are GET.

Endpoints

POST /v1/search — search authorities
GET /v1/research — research over the corpora
GET /v1/statute — retrieve a statute section by citation
GET /v1/opinion/{id}/text — full opinion text
GET /v1/cluster/{id}/text — full cluster text
POST /v1/citation-lookup — resolve a citation
POST /v1/citing — find citing authorities
POST /v1/citation-count — citation counts