26. Claude Desktop
What it is: an extension that lets an AI chat app, such as Claude Desktop, use the timegrapher: read your recordings, record the watch in each position, and show the readings as charts in the conversation.
Why you’d use it: to ask about a take in plain words (“What do these figures say?”, “Is the amplitude lower crown down?”) and have the answer come from what was measured, not a guess. The chat app can also guide a whole positions test, telling you how to place the watch before each take. It suits someone who would rather ask than read the tables.
The extension, timegrapher-mcp, runs the command line (chapter 25)
with --json and hands the chat app what it prints, so the figures in
the chat are the command line’s, and the app’s on the same sound and
settings.
Installing
The Mac disk image and the Windows zip each hold Timegrapher.mcpb in
their Claude Desktop folder. It holds both programs, so nothing else
needs installing.
- On Windows, extract the whole zip first.
- Double-click
Timegrapher.mcpb, or drag it into Claude Desktop’s window. Or, in Claude Desktop’s settings, choose Extensions, Advanced settings, then Install Extension… and pick it. - Choose Install. Claude Desktop says the extension isn’t signed; that is expected. The timegrapher’s tools appear in the chat’s tools menu.
- Start a chat with the Test My Watch prompt from the same menu, or just ask, giving a recording’s whole path.
The Linux zip has no extension. Any chat app that runs local MCP
(Model Context Protocol) servers (VS Code, Goose, LM Studio and others) can use
timegrapher-mcp from the Command Line folder: give it the
program’s full path as the command, with no arguments.
The MCP page gives the steps, and how to add it to
Claude Desktop by hand.
What It Can Do
The chat app asks you before it runs each of these.
| It can | Runs | Why you’d ask for it |
|---|---|---|
| Read one recording | analyze | The rate, amplitude and beat error of one position, and whether a regular beat was found. |
| Read a positions test | session | Each position’s readings and the figures that sum them up (M8), with a report page. |
| Read how the readings move | series | Steadiness: trends, level changes, cycles and how sure each is (chapter 9). |
| Read a long take | long | Rate and amplitude over hours, and the cycles that point at a wheel. |
| Show the Tick and Tock | profile, shape | When the amplitude or beat error seems wrong (chapter 7). |
| Record one position | record | A positions test, one take at a time. Uses the microphone. |
| Check the microphone | doctor | The level, clipping, automatic gain and noise. Uses the microphone, changes nothing. |
| Check a recording | doctor --file | The same, for a file. |
| Look up a calibre | calibres | Its beat rate, lift angle and wheels, your own additions included. |
| List your watches | watches | Each watch and its takes, with each take’s folder, so you can name a watch instead of a path. |
| List the sound inputs | devices | To pick the microphone. |
| Make a problem report | report | A zip to attach to an issue (chapter 27). |
| Read the guide | The watchmaker’s guide the chat app works from. |
It reads your watches while the app is open. Recording into a watch’s takes is refused while the app has the Watches folder open; close the app first.
The Test My Watch Prompt
The prompt, in the chat’s tools menu, starts the test an experienced watchmaker would run: the calibre, the microphone, then each of the six positions, recorded into one folder, and the result read as a positions test. Before each take the chat app says how to place the watch and waits for you. Each take is kept; when no regular beat was heard, or the sound clipped, the chat app is told so.
Charts
In chat apps that show MCP Apps (Claude on the web and desktop among them), five things show as charts in the conversation, drawn as the app draws them:
| Chart | What it shows |
|---|---|
| One recording | The rate, amplitude and beat error, those three over time, and the Tick and Tock. |
| A positions test | Each position’s readings and the figures that sum them up, with no pass or fail. |
| The Tick and Tock | Their profiles with the unlock and drop. |
| Recording | A Live view while it records: the readings, the last 30 seconds of the strip and the readings over time, about once a second, then what the check found. |
| Steadiness | The app’s Steadiness views, with each one’s headline and statistics, then the Isochronism walk and the readings’ distributions. |
One recording, a positions test, Recording and Steadiness list what limits the readings, when anything does, under What Limits These Readings. Chat apps without charts get the same figures as text.
What It Never Does
- Upload anything. A problem report is a file on your computer that you attach yourself.
- Change the computer’s sound settings. It reports what it measured; any change is yours.
- Use the microphone except to record a position or check the microphone, and only for the seconds asked.
- Grade the watch. It passes on what was measured; any judgement is the chat app’s, from the guide it reads, and yours.
When It Doesn’t Work
| What you see | What to do |
|---|---|
| “could not run timegrapher” | timegrapher isn’t beside timegrapher-mcp. Keep the two together, as the extension does. |
| “is not a full path” | The chat app passed a bare file name. Give it the whole path. |
| No timegrapher tools in the chat | Quit the chat app completely and open it again. For a server added by hand, check the path in its settings file. |
What It Can’t Tell You
- More than the command line does. Every figure comes from it; the chat app’s explanation is its own.
- That the chat app read the figures right. Its words are a model’s; the charts and the command line’s text are what was measured.
Previous: The Command Line · Next: Problems and Reports
This page comes from the app's user guide at v1.0.1. See it on GitHub.