Referent MCP server

Your firm's matters, in the AI apps you already use

Referent runs an MCP server at https://mcp.referent.law/mcp. Connect it in Claude, ChatGPT, Gemini, Perplexity, or any MCP-compatible agent, and that app can read and act on your firm's clients, matters, and documents. Same permissions, same approvals, same audit trail as the Referent app.

https://mcp.referent.law/mcp

The Referent MCP server is the endpoint - https://mcp.referent.law/mcp - that connects MCP-compatible AI apps to your firm's data in Referent. MCP, the Model Context Protocol, is the open standard that lets AI assistants use external tools; Claude, ChatGPT, Gemini, Perplexity, OpenClaw, Hermes, and a wide ecosystem of clients support it. Through the server, those apps read and act on your clients, matters, and documents under the same permissions, approvals, and audit trail you have in the app.

What you can do through it

Check a matter from wherever you already are

Ask Claude or ChatGPT for the current state of a matter - recent documents, correspondence, open items - without opening another tab. The answer comes from your live Referent workspace, scoped to what your account is allowed to see.

Ask what is due this week

Referent's deadline agent tracks hearings and deadlines and syncs both ways with your calendar. Through the connector, any MCP app can read that list, so "what do I owe the court before Friday" gets an answer from the system of record, not a guess. More on the agents at /ai-agents-for-law-firms/.

Draft correspondence that knows the file

Ask your assistant to draft a client update or a reply grounded in the matter's documents and history. Anything client-facing still waits for your approval in Referent - the connector does not change that rule.

Pull billing-ready summaries in plain conversation

The billing-prep agent assembles time and activity into billing-ready form. From an MCP app, you can pull that work up, question it line by line, and send it back for changes before anything reaches a client.

Find documents by describing them

Describe the document you need - the engagement letter on a given matter, the latest draft of an agreement - and the app locates it through the server, within your permissions. No folder archaeology.

Give an autonomous agent a supervised back end

If you run OpenClaw or Hermes, point the agent at the Referent server instead of ad hoc scripts. It works against the firm's actual clients and matters, and every step it takes through the server is logged in the audit trail, with client-facing and high-risk actions held for your approval.

Connect your AI app

Claude Claude
  1. In Claude's connector settings, choose to add a custom connector.
  2. Paste https://mcp.referent.law/mcp as the server URL.
  3. Complete the sign-in with your Referent account when prompted.
  4. Ask Claude about one of your matters to confirm the connection.
OpenAI ChatGPT
  1. In ChatGPT's connector settings, enable developer mode and choose to add a remote MCP server.
  2. Enter https://mcp.referent.law/mcp as the server address.
  3. Sign in with your Referent account to authorize the connection.
Gemini Gemini
  1. In Gemini's connector or tool settings, add a remote MCP server.
  2. Enter https://mcp.referent.law/mcp.
  3. Sign in with your Referent account when the authorization prompt appears.
Perplexity Perplexity
  1. In Perplexity's connector settings, add a remote MCP server.
  2. Enter https://mcp.referent.law/mcp as the server URL.
  3. Complete the Referent sign-in to finish the connection.
OpenClaw OpenClaw
  1. Add https://mcp.referent.law/mcp to OpenClaw's MCP client configuration file.
  2. Restart the agent and complete the sign-in with your Referent account - unauthenticated requests are rejected with a 401.
  3. Confirm the Referent tools appear in the agent's tool list before you hand it work.
Hermes Agent Hermes Agent
  1. Add https://mcp.referent.law/mcp to the MCP client section of Hermes Agent's configuration.
  2. Sign in with your Referent account when the agent first connects.
  3. Verify the connection by asking about a matter you know, then supervise from there.

...and any other MCP-compatible app - point it at the server URL above and sign in. Running an open-source agent? See the lawyer-focused guides for /openclaw-for-lawyers/ and /hermes-agent-for-lawyers/.

The security model

Same permissions, not new ones

A connected app can see exactly what your Referent account can see - nothing more. Workspaces are isolated per firm, access is role-based, and the server rejects unauthenticated requests with a 401.

The lawyer still signs

Referent's trust model applies over MCP exactly as in the app: AI prepares, and the lawyer approves every client-facing or high-risk action. "Legacy software records the work - Referent executes it, and the lawyer signs." A connector does not create a way around that.

Every step lands in the audit trail

Actions taken through the connector are logged the same way as actions in the app. That matters professionally: lawyers stay responsible for supervising work product, and a complete record is what makes supervision practical.

The same security posture as the rest of Referent

AES-256 encryption at rest, TLS 1.2+ in transit, per-firm isolated workspaces, and no training on firm data. CASA verified; SOC 2 and ISO 27001 are in preparation. Full details at /security/.

Frequently asked questions

What is MCP?

MCP - the Model Context Protocol - is an open standard, originated by Anthropic in November 2024, that connects AI assistants to tools and data. An MCP server exposes what a system can do; an MCP client, such as Claude or ChatGPT, discovers those capabilities and uses them. Referent's server at https://mcp.referent.law/mcp is how those apps reach your firm's data.

Which AI apps work with the Referent MCP server?

Any MCP-compatible client. As of mid-2026 that includes Claude, ChatGPT, Gemini, and Perplexity, plus open-source agent frameworks such as OpenClaw and Hermes Agent. Exact menu names vary by app version, so look for connector or MCP settings in your app and use the URL above.

Is my client data safe?

The server requires signing in with your Referent account over OAuth, and unauthenticated requests get a 401. A connected app operates under your account's permissions, inside the same controls as the app: AES-256 at rest, TLS 1.2+, per-firm isolated workspaces, no training on firm data (details at /security/). One general principle still holds - confidentiality obligations apply to any tool that touches client data, so the AI app on the other end of the connection is a choice you should make deliberately too.

Do I need to be a Referent beta member to use it?

Yes - the server requires a Referent account, and Referent is currently in beta. Cohort 1 is full: 450+ applications in under four weeks for 20 seats, with onboarding starting August 10, 2026 (the numbers are at /blog/cohort-1-full-what-450-applications-tell-us/). The Cohort 2 waitlist is open at /apply/, and every applicant gets founding-firm perks during the open beta.

What can the AI apps actually do through the connector?

They can read and act on your firm's clients, matters, and documents. Reading is direct: matter status, deadlines, documents, correspondence. Acting is bounded by the same trust model as the app - AI prepares the work, the lawyer approves every client-facing or high-risk action, and every step is logged in the audit trail.

Can I connect an autonomous agent like OpenClaw or Hermes?

Yes. Both are MCP-compatible, and both are serious tools - OpenClaw is the fastest-growing open-source project in GitHub history, with 347,000+ stars as of April 2026, and Hermes Agent from Nous Research shipped its desktop app in June 2026. Out of the box, though, neither has legal guardrails, an approval workflow, or an audit trail, and self-hosting means you own confidentiality and supervision. Connecting them through the Referent server adds that layer for work on your firm's clients and matters: the agent can prepare work around the clock, but client-facing and high-risk actions still wait for your sign-off, and every action through the server is logged.

Get a Referent account

The MCP server is part of Referent, and Cohort 1 of the beta is full. The Cohort 2 waitlist is open at /apply/ - every applicant gets founding-firm perks during the open beta. If you are still weighing what an AI-native practice looks like, start at /ai-native/.

Join the Cohort 2 waitlist