Claude for Legal Work: 7 Use Cases That Are Live Today
Lupl teamed with Midpage to present Claude use cases for legal practitioners. Not to debate whether AI matters for legal. That argument is largely settled. The question everyone was actually asking: what does it do, and what does it do right now?
We ran through seven use cases live, on real matter data, using real tools. Lupl connects to Claude via MCP. Midpage connects the same way, giving Claude access to every US case, statute, and regulation (the same corpus as Lexis or Westlaw), without leaving the chat. The two integrations are independent but work well together.
Why legal AI needs a system of record
The industries moving fastest on AI all made the same investment first. Engineering teams have Jira. Sales teams have Salesforce. Structured, live data is what gives an agent something real to work with: not just a chat window, but an actual view of what’s happening across the whole portfolio of work.
Legal hasn’t had that. Most firms are still managing matter plans in Word tables and tracking deadlines in Excel. Put a chat assistant on top of that and you can get some drafting help. You still can’t automate processes, and you can’t give an agent the context to be genuinely useful across a matter’s whole lifecycle. Lupl is that missing layer.
Seven use cases running today
1. Precision legal research
Midpage connects Claude to every US case, statute, and regulation (the same corpus as Lexis or Westlaw) inside the Claude chat interface. The difference from using Claude without it is grounding: no hallucinated citations, no cases pulled from training memory, real pin sites with clickable links to the full text.
The demo used a deliberately specific scenario: a slip-and-fall where the plaintiff tripped not on a wet floor, but on the wet floor sign itself. Keywords in a traditional tool surface hundreds of loosely related results. Claude with Midpage ran a multi-step search, read the closest cases, followed citations to better ones, and returned only what actually matched.
What used to mean opening and skimming dozens of cases now takes minutes. The citations feed directly into whatever comes next.
2. Cite checking
Upload a document (your own brief or one from opposing counsel) and ask Claude to check every citation. With Midpage connected, Claude reads the referenced cases, checks they exist, verifies that quoted passages appear in the cited source, and flags anything that’s been mischaracterised.
It’s the final-review step that’s easy to skip when time is short. Done in Claude, it takes a fraction of the time of a manual check.
3. Matter setup: from instruction to live matter in minutes
A client sends a new instruction: a dispute arising from a content licensing agreement. Drop the document into Claude, ask it to set up a matter in Lupl.
The agent reads the contract, identifies the governing law and arbitration clause (Singapore/SIAC, in this case), searches Lupl’s template library for the matching playbook, and builds out the full matter (workstreams, tasks, procedural steps) directly in Lupl. No clicks in the system. No form-filling.
A senior associate would typically spend 30–60 minutes on this. It ran in under two minutes. The agent didn’t just create tasks; it reasoned about which template fit, applied the matter context, and only asked for clarification where it genuinely needed it.
“The document had a Singapore arbitration clause. The agent found it, matched the right template, and scaffolded the whole matter around it.”
From there, the next steps come just as fast: draft the engagement letter, prepare a term sheet, run a conflicts check, all without leaving the interface.
4. Fee analysis
Pricing legal work is still mostly a partner’s instinct and a spreadsheet someone built years ago. This use case builds a live, interactive dashboard for the specific matter instead.
Ask for a fee quote. Claude pulls the matter context from Lupl and a pricing skill built on the firm’s rate cards, then produces a dashboard comparing fee structures (fixed, capped, hourly) side by side. Adjusting a discount slider updates margins across all options in real time. The agent flags its recommended approach and breaks costs down by phase and team level.
The output is built for the matter in front of you. Scope changes? You update the inputs. Client wants a different range? You ask for it. No rebuilding a spreadsheet.
5. Team capacity
Once a matter is set up, it needs to be staffed. Who actually has room for new work right now? Without a live view of task load across active matters, that question usually gets answered by asking around, and it isn’t reliable.
Ask Claude to build a capacity dashboard. It pulls live task data from Lupl (who has what assigned, how heavy their load is) and renders it as a visual overview. In the session it was immediately clear two team members were at capacity while others had room. That kind of clarity, without a team meeting.
6. Multi-jurisdictional tracking
Cross-border regulatory work means gathering advice from local counsel across multiple jurisdictions, in different formats, at different times. Tracking it and presenting it clearly usually means a complex manual tracker and a lot of assembly time.
Ask Claude to pull all local counsel advice from a matter in Lupl and map it visually: green for jurisdictions where advice is clear, red where there’s an issue or an outstanding question. Zoom in for detail, export as a PDF for the client package.
It works the same way for domestic state-by-state regulatory work. The map is built from actual matter data in Lupl, not a pre-built report with fixed fields.
7. Client portfolio reporting
When a general counsel calls to ask where things stand across all their matters, the usual answer is a long email or a scramble across multiple systems.
Ask Claude to build a client dashboard. It searches Lupl for all matters linked to that client, assesses status, flags blockers, and builds a visual overview from live data. Export it as a PDF and send it, or fold it into a regular reporting cycle.
“It’s so much better than a summary email. The client gets a real view of where everything stands, without the firm having to build it manually every time.”
Skills: how firm knowledge gets into the agent
Every use case above ran in part on a skill: a plain-English instruction file that tells the agent how to handle a specific task. A matter setup skill. A pricing skill. A capacity dashboard skill.
Skills are written by people (a knowledge management team, a senior lawyer who knows how work should be done) and deployed firm-wide. Fee earners don’t write prompts. They just ask, and the agent follows the firm’s process. Skills can also reference templates and firm branding, so the output looks like it came from your firm rather than a generic AI.
They’re shareable with one click, and they’re an open standard, so anything built today works regardless of which AI tool the firm is using six months from now.
A note on security
Claude’s Teams and Enterprise plans include zero data training, SOC 2 and ISO certifications, and data processing agreements. These are meaningfully different from the consumer version of Claude, which we’d recommend against using for legal work.
Midpage deletes all query data immediately. No matter content is stored. Lupl’s MCP integration is OAuth-based, so Claude can only see what the authenticated user is already permitted to see in Lupl. The security model doesn’t change; it extends into the AI layer.
The bottom line
These use cases ran live, on real data, in real time. All of them are available now, on infrastructure that exists today.
The firms making the most progress aren’t waiting for a single purpose-built legal AI to solve everything. They’re connecting the AI they already have to the systems that hold their data, and letting it work. Legal has always been process-heavy. That turns out to be useful.
