Matter summaries and handoff notes
Claude handles long material well. Give it a matter’s notes, emails, and documents and ask for a handoff summary with references a reviewer can check line by line.
Claude for lawyers
Claude reads long records and drafts carefully, but only from what it can see. FirmOps, a Vertana brand, builds and runs an MCP server that connects your case system (Clio, Filevine, SmartAdvocate, others), Gmail or Microsoft 365, documents, billing, and phones to Claude. The brain drafts and stages. A human always releases.
By Jonathan Mahler · Updated November 1, 2026
Practical uses
A personal-injury file is hundreds of pages of records, bills, and correspondence. That is the kind of material Claude summarizes well, as long as a person checks the output against the source.
Claude handles long material well. Give it a matter’s notes, emails, and documents and ask for a handoff summary with references a reviewer can check line by line.
First-pass chronologies from medical records: provider, date, complaint, treatment, gaps in care. Every entry should cite the page it came from, and a person corrects the draft before it is relied on.
Issue lists, missing-document checks, damages tables built from the bills on file, and draft structure. Claude does not set strategy, case value, or final language. A lawyer does.
Load the firm’s SOPs, templates, and AI policy into a project so staff can ask how the firm does something before they escalate the edge case to a manager.
Plans and confidentiality
The confidentiality question is mostly an account question. Here is how the plans split, based on Anthropic’s commercial terms. Read the current version yourself before you decide.
These run under Anthropic’s commercial terms, which say Anthropic “may not train models on Customer Content from Services.” Admins manage members and settings. This is where firm work belongs.
Consumer accounts run under consumer terms with their own data and training settings, which the user controls, not the firm. Check the current settings before anyone uses one, and keep client facts out of them.
Whether firm data trains anything, how long it is kept, who in your workspace can see it, and which connectors are allowed. Keep the answers with your AI policy.
The ethics frame is ABA Formal Opinion 512 (July 2024): competence, confidentiality, client communication, supervision of lawyers and nonlawyers, and reasonable fees. Sourced lines are on our law firm AI statistics page, and a law firm AI policy turns them into rules staff follow.
The limit
Ask Claude about a client and it has nothing to go on. It cannot see your case system, your inbox, your document storage, or your billing. So staff do the obvious thing: they copy notes and records into a chat. That is shadow AI. There is no record of what went in, no limit on who pasted what, and no review step before the output is used.
The fix is not a ban. It is a connection the firm controls, so Claude reads the file in place with the right permissions and every draft waits for a person.
Connecting Claude
MCP, the Model Context Protocol, is the open standard Anthropic introduced for connecting AI apps to other systems. A system exposes what it can do through an MCP server, and Claude uses it through one controlled door. The law-firm explainer is what is MCP for a law firm.
A project holds instructions and reference documents for a body of work: your templates, your style rules, your intake script. Useful for firm knowledge. It is not a live feed from your case system.
Claude offers connectors for common tools, such as email, calendars, and cloud drives. Availability depends on plan and admin settings and changes often, so confirm what your workspace allows before you plan around it.
Claude can connect to remote MCP servers as custom connectors, and the Claude desktop app can run MCP servers on the computer. That is how a case system, billing system, or phone platform gets in when no ready-made connector exists.
One MCP server for your firm across the case system, Gmail or Microsoft 365, documents, billing, and phones. Scoped by role, logged, and built around your fields and stages. Claude reads through it and prepares work: drafts in email drafts, proposed notes and task changes behind approve and reject controls.
The brain drafts and stages. A human always releases. The same server works with ChatGPT, so the firm is not tied to one model.
We build it and run it. The brain is yours. More on the Managed Firm Brain, the systems we connect, and how it sits next to Clio Work and Clio Manage AI.
Example prompts
Once the systems are connected, these are ordinary questions. The answers come back with sources, and anything that would leave the firm waits for a person.
| You ask Claude | The brain does | A person releases |
|---|---|---|
| “What’s the status of the Garcia matter?” | Reads the matter, recent notes and tasks, open requests, and the last client contact, and answers with sources. | A read. Nothing leaves the firm. |
| “Build a treatment chronology for Garcia from the records folder.” | Reads the records in the document system and drafts a dated chronology with page references and flagged gaps in care. | A paralegal checks each entry against the records. |
| “Draft an update for every matter quiet over 30 days.” | Finds matters with no client contact in 30 days and stages a plain-language update for each in email drafts. | Staff review and send. |
| “Which matters have a deadline in the next 90 days and no demand drafted?” | Compares calendared dates with matter stage and documents, and lists the files at risk. | The lawyer decides priorities. Deadlines stay on the firm’s calendar of record. |
| “Compare the bills on file with the specials in the draft demand.” | Totals the bills in the file, lines them up against the draft, and lists mismatches. | The paralegal fixes the draft. A lawyer signs the demand. |
| “Summarize every note on this matter since the last client update.” | Pulls notes, emails from Gmail or Microsoft 365, and call logs since the last update and summarizes them. | Used internally, or turned into a client update a person sends. |
| “Which clients signed this month still lack a welcome packet?” | Checks signed intakes against sent documents and lists who is missing what. | Staff send the packets. |
| “What did we bill last month, by matter owner?” | Reads the billing system and returns totals with the underlying entries. | The owner reviews. No invoice goes out on the brain’s say-so. |
Chronologies, records follow-up, and packet checks are staff work first. See how that plays out in legal AI for paralegals.
Safeguards
Claude prepares. People decide. Live client calls stay with your team, and the brain works behind the phone.
FAQ
On a commercial account the firm controls, with written rules and human review, yes for drafting and summary work. Anthropic’s commercial terms say it may not train models on customer content from those services. Personal consumer accounts have separate settings and should not hold client information.
Through MCP. Claude supports custom connectors that point at remote MCP servers, and the desktop app can run MCP servers locally. A case system with your firm’s fields, permissions, and approval rules needs a server built for it. That is the part FirmOps builds and runs.
Anthropic introduced the Model Context Protocol as an open standard in November 2024. ChatGPT and other AI apps now support it too, so a connection built on MCP is not locked to one model.
It is a strong first-pass tool because it handles long documents well. Treat the output as a draft: require page references for every entry and have a person check it before anyone relies on it.
It depends on the use. ABA Formal Opinion 512 says informed consent can be required when client information goes into a generative AI tool, and boilerplate engagement-letter language may not be enough. Check your state rules. This is general information, not legal advice.
Both work. Many firms prefer Claude for long documents and careful summaries, and ChatGPT where staff already live in it. Build the connection layer so it works with either, and switch when the next model is better.
No. Your team stays in Claude (or ChatGPT). FirmOps, a Vertana brand, builds and runs the MCP server that connects your case system, email, documents, billing, and phones, and keeps every outgoing action behind a person.
Still choosing a model? Read Claude vs ChatGPT for lawyers, or why we tell firms to live in Claude or ChatGPT rather than adding another legal AI tab.
Keep reading
The firm brain: Claude with your case system, email, and files connected.
Plans, ethics duties, and connecting ChatGPT to the case file.
Choose by review burden, not model hype.
What AI does in a firm today, by function, and how to start.
The categories, the named tools, and where a firm brain fits.
Clio Work, Clio Manage AI, and the rest of your stack.
Jobs inside one brain, not a catalog to babysit.
Primary-sourced numbers on adoption, ethics, and time saved.
Start safe, prove it with staff, then expand.
Next step
In a 30-minute live demo we run a firm brain against one real job, map your stack, and confirm your systems have the APIs to connect.
Put a number on it
Claude Team is a line that stays. The point tools around it are the question. Check with the law firm software cost calculator. It starts from a real firm's stack, tags each line Stays or Consolidates, and nets the $500 a month FirmOps maintenance; what that covers is on the pricing page.