# MCP Server Documentation - Relex > Relex MCP Server — give your AI agent access to the entire Relex Legal AI API with just two tools (search and execute) in ~1,000 tokens. [Home](/) / [Docs](/docs/a2a) / MCP Server ### A2A Protocol - [Overview](/docs/a2a) ### MCP Server - [Overview](/docs/mcp/) - [Quickstart](/docs/mcp/quickstart/) #### On this page - [What the agent does — and what it never does](#clause-2) - [Two tools](#clause-3) - [Clause enforceability — agree with Relex before you rely on a clause](#clause-4) - [The PII boundary](#clause-5) - [Connect](#clause-6) - [Typical loop](#clause-7) # Relex MCP Server The Relex MCP (Model Context Protocol) server lets a reasoning agent — Claude (Claude Code or Claude desktop) or any MCP client — **work on your Relex cases** through two tools and about **1,000 tokens** of context, **without ever seeing** **your clients' personal data**. ## What the agent does — and what it never does The agent is **reasoning, not operations**. Over the API it can: - **Read** your cases, drafts, timeline, and structure — always de-identified (it sees labels like `[Party 1]`, never real names or IDs). - **Reason** about the matter and **draft** documents. - **Record** its work back to the case (notes, drafts, structure). The agent **cannot** add your know-how, create parties, upload documents, or handle personal data. Those are done by **you, in Relex**, in your browser — where personal data is extracted, encrypted with a password only you know, and stored encrypted. When something needs to be added, the agent **points you to** **the right place in Relex** instead of doing it itself. ## Two tools The whole API is reachable through two tools, so the context footprint stays fixed no matter how large the API is: | Tool | Arguments | Purpose | | ----------- | ------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **search** | { query?, tag?, method? } | Discover endpoints. Returns { method, path, summary, tags, params } — params shows each call's required/optional fields, enums and defaults, so the agent never guesses a field or asks you for one the API defaults. | | **execute** | { method, path, query?, body? } | Call an endpoint. The server validates it, enforces the PII guard, runs it with your auth, and returns { status, body }. | Both take **plain arguments** — there is no code to write. `path` is relative to `/v1` (for example `/cases`, `/cases/{id}`), with no percent-encoding. search({ query: "cases" }) execute({ method: "GET", path: "/onboarding/status" }) // Open a case — never set the tier or name yourself; Relex's eval agent decides // the tier from the matter. Create the case, then drive the eval agent: execute({ method: "POST", path: "/cases", body: {} }) execute({ method: "POST", path: "/agent", body: { type: "eval\_req", caseId: "", payload: { prompt: "" } } }) // The eval agent returns its question or the tier + offer; relay any question, // repeat until it returns the offer, then read the case back with GET /cases?caseId=. ## Clause enforceability — agree with Relex before you rely on a clause When you review or draft a clause-bearing document, **check enforceability with** **Relex** before treating any clause as sound. Relex's case agent grounds each verdict in the jurisdiction's mandatory law; you bring your own read; together you reach an explicit, question-driven agreement — and where you disagree, a human lawyer decides. search({ query: "enforceability" }) execute({ method: "POST", path: "/cases/{caseId}/draft/{draftId}/enforceability-advisory", body: { clauseText: "", clauseContext: "employment, governing law X", claudeVerdict: "unenforceable" } }) Relex returns its grounded `verdict` (enforceable / questionable / unenforceable / void), the deciding `doctrine`, a compliant alternative, and the **open questions** that must be answered first. When your verdict and Relex's **disagree** — or Relex's not-enforceable verdict is **ungrounded** — the response is `status: "needs_human"`: both verdicts are recorded as an internal comment, the clause is flagged, and **the** **human professional is the tiebreaker**. Neither agent auto-resolves; the gate's job is to _stop and ask_. A clause left flagged unenforceable/void blocks the draft from being sent for signature until a human resolves or overrides it. Send only de-identified clause text — the PII boundary below still applies. ## The PII boundary Personal data — names, national IDs, contact details, and document content — is **end-to-end encrypted** and only ever decrypted in your browser. The server **refuses** any call that would move plaintext PII (reading or writing parties, uploading or reading document content) and instead returns a **deep link** so you do that step securely in Relex. The agent works with de-identified labels and anonymized counts only. ## Connect The agent signs in over **OAuth** — there is no key to paste: 1. Install the plugin from GitHub: `/plugin marketplace add relexyou/relex-claude` then `/plugin install relex-legal@relex`. 2. On the first tool call, your browser opens to sign in to Relex (Google or Apple) and approve. A per-connection credential is provisioned automatically. In Claude desktop / claude.ai, add a custom connector with the URL `https://relex.you/api/mcp`, then click **Connect** to sign in. On **Pro / Max** you add it yourself under **Customize → Connectors → Add custom connector**; on **Team / Enterprise** an owner or admin adds it once in the **organisation's** **settings → Connectors**, and each member connects it from **Customize →** **Connectors → available connectors**. A static API key from **Settings → API Keys** is supported as a CI/headless fallback. See the [Quickstart](/docs/mcp/quickstart/) for the exact steps and the prompt to paste. ## Typical loop | You want to… | The agent… | | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Start a matter | execute POST /cases, then POST /agent eval\_req with the matter — Relex's eval agent sets the tier; on payment it deep-links you to billing. The agent never asks you the tier or quotes a price. | | Bring in the people/documents | points you into Relex to add parties and upload documents (encrypted in your browser) | | Understand the matter | reads the case, timeline, and drafts (de-identified) and reasons over them | | Produce or fix a draft | drafts/redlines in your style, grounded in your know-how, and records it via POST /cases/{id}/draft | | Capture a firm position | builds a non-confidential playbook (POST /playbooks) the redline path reuses | | Export with real details | points you to Relex, where the document is re-identified locally | --- _[View this page on relex.you](https://relex.you/docs/mcp)_