Clio Work vs Claude for Law Firms: What Each Does and Where Each Falls Short

Law firms that use Clio and have a Claude subscription keep running into the same question: what is Clio Work for, and when do you use Claude instead? The two products look like they overlap, and in certain ways they do. Firms that have tried both tend to end up with a fairly specific split between them, and understanding why requires knowing what each product actually does.

This article covers what Clio Work and Claude each handle well, where each runs into limits, and a third setup that some firms add when they want Claude's reasoning applied to their actual Clio data.

---

What Clio Work does

Clio Work is Clio's AI workspace, built into Clio Manage and available as a standalone product since April 2026. The core function is AI-assisted Q&A and multi-step task completion within the Clio platform. You can ask it questions about your matters, have it pull together summaries from Clio records, and in its agentic mode, have it complete sequences of steps without switching between tools.

The situations where Clio Work fits cleanly:

• You want a quick answer while you are already working inside Clio and you do not want to switch applications.
• Your question involves a specific matter or client that Clio Work can see through your account.
• You need a task completed that Clio has specifically built the tool to handle within its own workflow.

Where it runs into limits: the model quality for complex writing and reasoning tasks is lower than Claude, which is what sends some attorneys and paralegals back to Claude when the work requires precision. Clio Work also operates entirely inside the Clio interface. It does not participate in workflows that live outside Clio, and it does not give you programmatic access to your Clio data from other tools.

---

What Claude does

Claude (Anthropic) is a general-purpose AI with strong writing and reasoning. For legal work the tasks it handles well: drafting motions and letters, reviewing and summarizing documents, analyzing facts against legal standards, preparing questions for depositions or client calls, and working through ambiguous legal arguments. The model quality for extended, precise writing is what draws most lawyers to it over built-in practice management AI tools.

The gap: Claude has no access to your firm's data. Every session starts blank. To ask Claude about a specific matter, client, or billing record, you need to paste that information into the conversation. For one-off tasks that is manageable. For workflows that require the same Clio context every time, the copy-paste becomes the bottleneck.

Two things worth knowing on the confidentiality side:

• Claude Team excludes your conversations from Anthropic's training. That covers the "is my client data being used to train AI" concern.
• Claude Enterprise adds Zero Data Retention (ZDR): content is not stored on Anthropic's side after inference completes. This is the privilege-relevant control for firms sending privileged communications to a hosted model.

With any hosted AI, the content you send reaches the model for inference. That is the access event. ZDR governs what happens to it afterward. The two are different questions, and conflating them leads to claims that won't hold up to review.

---

Where neither tool fully covers the automation use case

The pattern that doesn't work cleanly with either tool on its own: a recurring workflow that reads your Clio data, applies some reasoning, and writes the result back into Clio or another system.

An example: after every matter intake, someone on your team drafts a welcome email that pulls the client's name, the matter type, and the assigned attorney from Clio. With Clio Work, you can do some of this inside the Clio interface, but the outputs stay inside Clio. With Claude, you can draft a better email, but you need to manually paste the client name, matter type, and attorney every time.

The same gap appears with billing narration (you want Claude to write descriptions from your time entries, but Claude doesn't see your time entries), pre-call summaries (you want Claude to pull together the matter history, last contacts, and open tasks before a call), and document-triggered tasks (you want Claude to create a Clio task after reviewing a document, but Claude and Clio don't talk to each other by default).

---

The Clio MCP connector

The Clio MCP connector (@oktopeak/clio-mcp on npm, MIT license) is an open-source Model Context Protocol server that gives Claude direct read/write access to your Clio account. It runs as a local server on your machine, authenticates with your Clio API key, and surfaces your Clio data as tools Claude can call inside a conversation or an automated workflow.

With the connector running, Claude can:

• Search clients and matters by name, matter number, or status
• Read billing entries, time logs, and invoice status for a matter
• Pull calendar events and deadlines for a specified range
• Read and create tasks on a specific matter
• Search and retrieve documents from a matter's file
• Access contacts and their relationship to matters

Every tool call that reads or writes Clio data writes a timestamped entry to a local append-only audit log. The log records the timestamp, the tool name, which client or matter was accessed, and your Clio user ID. This is the electronic communication security log that ABA Opinion 512 (2023) asks for when AI tools handle client communications.

The connector runs on your machine. Your Clio API credentials are stored in an encrypted file locally. No data routes through Oktopeak infrastructure. The MCP server is open-source code you run.

---

What this changes in practice

Here are three workflows that become cleaner with the connector in place.

Billing narration. Ask Claude to pull last week's time entries for a specific matter and write billing descriptions ready to paste into Clio. Claude calls the billing tool, gets the raw entries, writes the descriptions. The alternative is exporting a CSV and pasting it into Claude every time.

Pre-call preparation. Before a client call, ask Claude to summarize the matter history, most recent correspondence, open tasks, and any billing items that might come up. Claude pulls the data from Clio directly and produces a two-paragraph summary. The alternative is opening four different Clio screens and synthesizing the information yourself.

Post-review task creation. After reviewing a document in Claude, tell Claude to create a follow-up task in Clio with the relevant matter and due date. Claude calls the task creation tool. The task appears in Clio. The alternative is switching to Clio and manually entering the task after finishing the review.

These are not tasks Clio Work handles programmatically from outside the Clio interface, and they are not tasks Claude handles without Clio data access.

---

Privilege and ABA Opinion 512

ABA Formal Opinion 512 (2023) addresses what "reasonable care" looks like when AI tools process client communications. The requirements are not prescriptive about specific technology, but they focus on security of the communication channel, understanding what the AI system does with the content, and having appropriate supervision of AI-generated output.

The Clio MCP connector's local audit log is designed to address the Opinion 512 security requirement. The data path: your Clio data travels from Clio's API to your local machine, and from there to the Claude model for inference. Nothing goes through a third-party server in between.

The AI provider side is separate: with Claude Enterprise and ZDR, conversation content is not retained after inference completes. Anthropic's data processing agreement covers the enterprise tier. If you are sending privileged client communications to Claude, you need a signed BAA on the Anthropic side in addition to the local connector controls. The connector addresses the local handling; the Anthropic agreement addresses what happens at the model.

For a detailed breakdown of the access-vs-retention distinction and the full BAA + ZDR + audit-log stack, see our post on how the compliance architecture of the Clio MCP connector works.

---

Which setup fits which situation

Some practices run all three. Clio Work for quick queries inside the Clio interface. Claude for heavier document work and drafting. The Clio MCP connector for the specific workflows that need live Clio data access.

---

Setup and next steps

The Clio MCP connector source code is at github.com/oktopeak/clio-mcp. Installation runs through npm. The README covers the full setup process including the Claude Desktop config snippet and how to verify the audit log is working.

If you want to work through which Clio workflows would benefit most from Claude access before spending time on setup, a 30-minute discovery call covers your current practice setup and identifies the top two or three automations worth configuring first.