TextExpander MCP prompts: how to clean up your Snippet library with Claude

A shared Snippet library is one of the most useful things a team can build. The right answer gets written once, everyone who types the abbreviation sends that same answer, and when something changes, one edit updates it for the whole team.

However, like any knowledge base, a library that grows for years needs a little upkeep. Products change, links move, and a few near duplicates creep in. Reviewing hundreds of Snippets by hand takes time, and that’s where the TextExpander MCP Server helps. It connects an AI assistant like Claude to your library, so Claude can read every Snippet, flag what needs attention, and make the changes you approve.

TextExpander’s own support team used it to review 862 Snippets across 16 groups. If your library is ready for a refresh, its workflow and the prompts that accompany it are a practical place to start.

What the TextExpander MCP Server can do

MCP stands for Model Context Protocol, an open standard that lets AI assistants use tools in other apps. The TextExpander MCP Server is TextExpander’s connector for it. According to the TextExpander MCP Server Getting Started Guide, as of November 2026 it can:

• Read your library. List Snippet Groups, show the Snippets inside them, and search by abbreviation, label, or content.
• Report usage. Pull Snippet statistics, including expansion counts, characters saved, and activity.
• Create. Add new Snippets and Snippet Groups, including Fill-ins and dates written as plain-language descriptions.
• Update and move. Change a Snippet’s content, abbreviation, or label, and move a Snippet to another group.
• Delete. Remove a Snippet or a whole group. Deletion is permanent, and the assistant asks you to confirm before anything is removed.

It works with Claude Desktop, Claude with Cowork, ChatGPT in Developer Mode, and Microsoft Copilot through Copilot Studio. Permissions carry over from TextExpander. The assistant can see only the Snippets and groups your account can see, and people without edit rights can’t edit shared Snippets through it.

Most of this post uses Claude, because that’s what TextExpander’s support team used. The same prompts work as a starting point in the other clients. For setup steps, follow the Getting Started Guide.

How TextExpander’s support team cleaned up 862 Snippets

Greta Jensen leads Customer Experience at TextExpander, and her support team runs on a shared Snippet library. Over the years, that library grew to 862 Snippets in 16 groups. Some answers described features that had changed, and others pointed to pages that no longer existed.

Three members of the team worked through the library group by group with Claude connected to the TextExpander MCP Server. They used it for the inventory, 12 months of expansion counts, dead-link scans, and before-and-after comparisons. In the end, they:

• Deleted 391 outdated or redundant Snippets.
• Archived 58 Snippets the team might need again.
• Edited or updated 196 or more.
• Resolved 16 duplicate pairs or sets.

The review also caught problems nobody had been looking for, including an outdated list of supported languages, references to vendors the team no longer used, and stale URLs.

The MCP made a measurable difference. In Greta’s notes, one teammate compared the first five groups she cleaned up without the MCP to those she cleaned up with it. The manual groups took longer and were less accurate.

Usage data turned “what should we cut” into an evidence-based call instead of a guess.

Greta Jensen, Head of Customer Experience, TextExpander

Before you start: five rules for a safe cleanup

These rules come from what the support team learned the hard way. Each one prevents a mistake that’s easy to make when an AI assistant can edit your library.

1. Start read-only. Ask Claude to report before you ask it to change anything. Every prompt in the audit stage ends with “Don’t change anything.”
2. Work one group at a time. Large groups may be more than Claude can process in a single pass. When Claude can’t finish a step, ask it to split the group into smaller categories and work through those.
3. Save a baseline export from the TextExpander website. The team found that exports pulled from the TextExpander website held up better than AI-generated exports. Compare a fresh website export with the baseline when you’re done.
4. Treat usage data as a signal. A Snippet with zero expansions may be embedded inside other Snippets, and a Snippet with low personal use may still be a high-use team Snippet. Confirm every deletion by hand.
5. Check a few Snippets after every edit. Open some of the changed Snippets in TextExpander and confirm the formatting, Fill-ins, and links still look right.

Usage data is a strong signal, not a verdict, which is why we always confirm manually before deleting.

Greta Jensen, Head of Customer Experience, TextExpander

Keep customer data out of the process as well. Shared Snippets should hold approved wording, not account details or patient information, and the prompts you paste into Claude should follow the same rule.

A seven-step Snippet cleanup workflow with Claude

Greta combined the three teammates’ approaches into a single repeatable workflow. Replace [Group name] with the group you’re reviewing, and run the steps in order.

1. Take an inventory and save a baseline

Get a count of Snippets in each group and a “before” snapshot, so you can confirm changes later instead of relying on memory. For large groups, export the group from the TextExpander website and give that file to Claude to work from.

Prompt: “List every Snippet in the [Group name] group with its label, abbreviation, and last edited date, and export it as a CSV.”

2. Pull usage data

Expansion counts turn “what should we cut” into a decision you can explain.

Prompt: “Pull expansion counts for every Snippet in [Group name] over the last 12 months and merge them into the CSV.”

3. Flag candidates without deleting anything

Look for broken links, zero-use Snippets, and near duplicates. Greta’s workflow splits this into two prompts. 

The first checks links and usage, and confirms that a zero-use Snippet isn’t embedded in another Snippet before flagging it. The second compares Snippets with each other to find duplicates, which is a different kind of check and easier for Claude to handle in its own pass, especially in a large group.

Links and usage prompt: “Check every URL in [Group name] for broken or dead links, flag any Snippet with 0 expansions in the last year, and check whether any of those zero-use Snippets are embedded inside other Snippets before flagging them for deletion.”

Duplicates prompt: “Compare all Snippets in [Group name] and flag any that are near duplicates or that produce the same or similar output.”

4. Build a tracking sheet

Keep one running table with a status, a reason, and notes for each Snippet. Logging decisions as you go makes it easier to correct course halfway through or pick the work back up later.

Prompt: “Build a tracking table with columns for Snippet name, status (keep, edit, archive, delete), reason, and notes, based on the flags above.”

5. Review every deletion by hand

This is the one step nobody should skip. Read each flagged Snippet and its history before you confirm.

Prompt: “For each Snippet flagged ‘delete,’ show me its content and expansion history before I confirm.”

6. Make the changes, and archive before you delete

Apply the edits, merges, and deletions you approved. If a Snippet might be useful again, like an answer for an older version of your product or a policy you could bring back, copy it to an archive document before you delete it, in the same step. Use the first prompt for edits and the second for anything you archive and remove.

Edit prompt: “Update the [Snippet name] Snippet’s content to [new text].”

Archive prompt: “Copy [Snippet name] to our archive document, then delete it from [Group name].”

7. Compare before and after

Export a fresh copy of the group from the TextExpander website, then have Claude compare it to your baseline. The result is a verified summary of what changed.

Prompt: “Compare this new export of [Group name] with the original CSV and list exactly what changed.”

TextExpander MCP prompts for regular library checks

A big cleanup resets the library. Smaller checks keep it from drifting again. These prompts are written for a library owner, and each one either stays read-only or waits for your approval.

Audit prompts (read-only)

• Duplicates: “Search my Snippet Groups for Snippets with the same or nearly the same content. Show a table with the group, abbreviation, label, first line, and which one you’d keep and why. Don’t change anything.”
• Abbreviation conflicts: “Find abbreviations that are identical, or where one abbreviation is the start of another, like ;add and ;address. List each conflict with its groups. Don’t change anything.”
• Labels: “Find Snippets with a blank, duplicate, or unclear label. Suggest a clear label for each in a table. Don’t change anything.”
• Stale content: “Search my Snippets for content that may be out of date, including past years, these old names or terms [list them], and URLs. List each Snippet with what looks stale. Don’t change anything.”
• Compliance review: “Find every Snippet that mentions [words or phrases]. Return a table with group, abbreviation, label, and full content so I can send it for legal or compliance review. Don’t change anything.”

Fix prompts (with your approval)

• Replace outdated wording: “In my [Group name] group, show me every Snippet that contains [old text] and what it would look like with [new text]. Wait for my approval before you update anything.”
• Clean up abbreviations: “Propose a consistent abbreviation scheme for my [Group name] group. Show a before-and-after table and check that no new abbreviation conflicts with another group. Wait for my approval before you update anything.”
• Reorganize a large group: “My [Group name] group has grown too large. Suggest a better set of groups and which Snippets belong in each. Don’t change anything until I approve the plan.”

Gap prompts

• Missing answers: “Here are our most common customer questions this month: [paste list]. Compare them with my [Group name] group and list the questions that have no Snippet. Draft a Snippet for each, and wait for my approval before you create a new group called ‘Drafts for review.'”

Drafts in a review group still need the same check as any other new content. For a closer look at where people should review AI output, see our guide to human-in-the-loop AI workflows.

How to schedule a weekly library check with Claude scheduled tasks

Claude scheduled tasks run a saved prompt on a schedule, such as hourly, daily, weekly, or on weekdays. According to Anthropic’s help center, as of November 2026, they’re available on paid Claude plans, can use the connectors you’ve set up in Cowork, and run remotely, so they keep their schedule when your computer is asleep.

To set up a weekly library check in Claude Cowork:

1. Click Scheduled in the left sidebar, then click New task.
2. Choose Set up manually, and name the task something like Weekly Snippet library check.
3. Paste the weekly check prompt, choose Weekly, and save.

Keep the scheduled run read-only. A scheduled task runs when no one is watching, so it should report what it finds, and someone should make the fixes in a live session where they can review each change. If a run can’t reach your library, open Claude and sign in to the TextExpander connector again.

Send the weekly report to the person who owns the library. A short read each week can catch a new duplicate or an outdated term before it spreads across your team’s replies.

Where AI helps with a Snippet library, and where people stay in charge

The support team’s cleanup showed what AI does well with a library. It handles the inventory, compares hundreds of Snippets, scans links, and turns usage data into a shortlist. In Greta’s words, AI makes maintaining a Snippet library affordable.

It also showed where people need to stay involved.

• Recommendations need checking. During the cleanup, Snippets flagged as unused turned out to be embedded in other Snippets, and one Snippet flagged for deletion was a high-use team Snippet.
• AI-generated reports need verifying. Exports from the TextExpander website were more reliable than CSVs Claude generated, so the team compared real exports instead of trusting Claude’s own tracking.
• Big jobs need breaking up. Large groups were too much for one pass. Splitting them into categories became a normal part of the workflow.

That division of work is also why a Snippet library holds up so well alongside AI. Each Snippet is written once, reviewed once, and owned by someone, so a wrong answer can be found and fixed in one place.

A stale snippet gets found, owned, and fixed once. A wrong generated answer can only be caught one reply at a time.

Greta Jensen, Head of Customer Experience, TextExpander

Customers see the same benefit. CompanyCam’s onboarding team uses a shared library across several tools, so a single update reaches everyone.

I can update a TextExpander Snippet, and the team has access to the most up-to-date content immediately.

Alex Williams, Customer Onboarding Manager, CompanyCam

More TextExpander MCP Server use cases

Cleanup is one job. The same connection between Claude and your library supports others:

• Turn your best AI prompts into Snippets. Ask Claude to save a prompt you reuse as a Snippet with Fill-ins for the parts that change. Our guide to saving ChatGPT prompts shows the idea, and the AI prompt library guide covers how teams share them.
• Turn a verified answer into the next Snippet. When someone on the team writes and checks a new answer, Claude can save it as a Snippet in the right group, which is how the support team grows its library between cleanups.
• Build a group from a document. Paste in a policy, a template set, or a list of standard replies, and ask Claude to create a Snippet Group from it for your review.
• Bring saved replies together. If your team keeps canned responses across several tools, Claude can help you consolidate them into a single shared group. See how teams manage saved replies across tools.

Keep the library worth trusting

A Snippet library is only as useful as its least accurate Snippet. The TextExpander MCP Server makes the review that keeps it accurate a weekly habit instead of a once-a-year project, and a person still approves every change.

Start with one group and the read-only audit prompts. When you’re ready, try TextExpander free and connect it to Claude, or explore more AI prompts you can save as Snippets.