Running Your Practice from Claude and ChatGPT: A Guide to MatterOS over MCP

What MCP means, concretely

MCP is a protocol, not a MatterOS invention. Anthropic published it as an open standard for letting an AI assistant call defined tools against a piece of software, the same way a plugin or an API integration would, except the assistant decides when to call which tool based on what you asked it in plain language. If you've connected a calendar or a file-storage app to Claude or ChatGPT before, you've used MCP already.

For MatterOS, connecting takes one URL (https://thematteros.com/mcp) pasted into ChatGPT's or Claude's connector settings, followed by an ordinary OAuth login - the same MatterOS email or Google login used in the app. There's no API key to generate, copy, or accidentally commit to a repo.

What a connected assistant can actually do

The tool list is public on the /connect page itself and grouped by pillar, matching the product's own structure rather than a generic chatbot's grab-bag of actions:

• Matters - list matters and filter by status, open one matter's full record, or pull a whole matter brief (parties, tasks, deadlines, findings) in a single call.
• Actions - pull the agenda of what's overdue, due soon, and coming up across every matter; list, create, and update tasks with urgency and due dates.
• Team - list the people on a matter (clients, opposing parties, witnesses, counsel) and add a new party or contact.
• Evidence - find documents by matter, title, or tag.
• Research - list findings (facts, issues, risks, strategy, ranked) and add a new finding or flag.
• Time - pull the chronology with logged and billable minutes, and log time directly onto a matter.

What that looks like as an actual conversation

The point of MCP isn't that these actions are impossible in the app - it's that they no longer require opening it. Four real examples, each mapped to a tool that ships today:

• "What's overdue or due this week across all my matters?" calls get_agenda and returns a ranked triage list without opening a dashboard.
• "Brief me on the Smith v. Jones matter." calls get_matter_brief - parties, tasks, deadlines, and findings in one shot, useful right before a client call.
• "Add a high-urgency task on Smith v. Jones to draft the motion, due Friday." calls create_task and puts the deadline directly on the matter, no re-typing it into the app afterward.
• "Log 45 minutes on the Wynn call, billable." calls log_time, so a call that ends with the phone already in your hand doesn't turn into a lost half-hour of unbilled time.

How the scoping and safety actually work

Every call runs as you, not as a generic service account. When you connect, you approve access through an ordinary browser OAuth consent screen using your own MatterOS login; the assistant then receives a token scoped to your permissions specifically. It inherits your row-level security, so it can never see another user's matters, tasks, or clients, no matter how it's asked.

Most of the 14 tools are read-only. The handful that write - create or update a task, log time, add a finding, add a person - are marked as writes in the tool list itself, so nothing changes in your matter file silently. Disconnecting the assistant from your connector settings kills the token immediately; nothing lingers with standing access.

Why this matters beyond convenience

Most practice management platforms treat AI as a feature inside their own app: a chat panel bolted onto the dashboard, closed to any assistant but the vendor's own. MCP inverts that - MatterOS's data model is available to whichever assistant you already trust and already use for everything else, rather than asking you to context-switch into a bespoke, vendor-locked chat window every time you want an AI's help with a matter.

For a solo or small-firm lawyer already living inside Claude or ChatGPT for drafting, research, or correspondence, that means the practice-management layer comes to the conversation instead of pulling you out of it. It's also, practically, a bet on where legal software is headed: the firms comfortable connecting their tools to an AI assistant today are the ones positioned to keep pace as more of the profession's software follows the same pattern.

Do I need a developer or IT person to connect MatterOS to Claude or ChatGPT?

No. Connecting is a copy-and-paste of one URL into the assistant's connector settings, followed by a normal browser login using your existing MatterOS credentials. It's designed for a solo attorney to do in under a minute, with no API key management.

Can my AI assistant see another attorney's matters at my firm?

No. The token your assistant receives is scoped to your account and inherits the same row-level security the app enforces - it can only see and change what you personally can see and change inside MatterOS.

Which tools can write to my matters, versus just read?

Most of the 14 tools are read-only: listing matters, tasks, people, documents, findings, and time. A small marked set can write - create_task, update_task, add_person, add_finding, and log_time - and each is clearly labeled as a write in the tool list on the /connect page.

What happens if I disconnect the assistant later?

The connection can be revoked at any time from the assistant's own connector settings. The token dies immediately, and the assistant loses all access to your MatterOS data from that point forward.

Does this work with both Claude and ChatGPT?

Yes. Both are MCP-capable: ChatGPT connects through its Connectors settings with Developer mode enabled, and Claude connects through its own custom connector settings. Any other MCP-capable assistant can use the same URL the same way.