Example prompts¶

Thirty-six prompts you can paste into your assistant, from a single look at a file to a complete calibration. Each one lists what the assistant will do, the tools it will call, the launch flags it needs and what you should check in the result.

Replace the names with your own

The examples use sample.mzML, standards, the folder C:\MSData and the compound caffeine as stand-ins. Change them to your own files, folders and masses. The assistant can only reach files inside a folder you gave with --allow-root; every prompt needs that flag, so it is not repeated in the lists below.

The assistant chooses its own steps, so the tools it calls can differ a little from the list. The list shows what the guide expects. In the window of chromascope-mcp, the link View log in the status bar opens the AI call log, which shows what really happened while the window is open.

Open and inspect data¶

1. See what is in a folder¶

Which mzML files are in C:\MSData? Give me the names and sizes.

The assistant will: list the files of the folder (not its subfolders).
Tools: list_mzml_files
Flags: none besides --allow-root.
Check: the names match what you see in the folder. Ask for include_vendor if you also want vendor files listed.

2. Open a file and show its TIC¶

Open sample.mzML and show me the TIC.

The assistant will: open the file, extract the total ion chromatogram and show it in the window.
Tools: open_files, extract_chromatogram
Flags: --allow-changes (without it the assistant can read the trace, but the window does not show it).
Check: the trace in the window is the TIC of the right file.

3. Describe a file¶

Open sample.mzML and tell me how many scans it has, the retention time range, the polarity and which MS levels are present.

The assistant will: open the file and read its metadata and a few scans.
Tools: open_files, dataset_metadata, list_scans
Flags: none besides --allow-root.
Check: compare the scan count and the time range with the file information in the window.

4. Open a whole folder and compare the files¶

Open every mzML file in C:\MSData\batch1 and give me a table with the file name, the number of scans and the retention time range.

The assistant will: list the folder, open the files (up to 64 in one call), and read each file's metadata.
Tools: list_mzml_files, open_files, list_datasets, dataset_metadata
Flags: none besides --allow-root.
Check: the number of rows equals the number of files; a file that failed to open is named with its error.

Chromatograms and spectra¶

5. Extract an ion chromatogram¶

In sample.mzML extract the XIC for m/z 258.1109 with 10 ppm tolerance, positive polarity, MS level 1, and show it.

The assistant will: extract the trace with exactly these settings and display it.
Tools: extract_chromatogram
Flags: --allow-changes to show it.
Check: the settings under Trace settings repeat your mass, tolerance and polarity.

6. Compare one mass in three files¶

Open run1.mzML, run2.mzML and run3.mzML. Extract the XIC for m/z 258.1109 at 10 ppm in each, and overlay them on one plot with a shared intensity scale.

The assistant will: open the files, extract the trace in each, switch to overlay and set a shared scale.
Tools: open_files, extract_chromatogram, set_layout, set_display
Flags: --allow-changes.
Check: three traces in one plot; the legend names the three files.

7. Look at the spectrum under a peak¶

In sample.mzML show me the mass spectrum at 3.2 minutes and tell me the five most intense peaks.

The assistant will: read the spectrum of the scan nearest that time, show it, and list the largest peaks.
Tools: spectrum, spectrum_image
Flags: --allow-changes to show the spectrum in the window.
Check: the retention time of the spectrum is what you asked for (it is the nearest scan, so it may differ by a few seconds).

8. Ask for a second opinion on a peak shape¶

Zoom in on the peak around 3.2 minutes in the XIC you just made. Does it look like it has a shoulder or a split top? Show me the picture you looked at.

The assistant will: zoom, take a picture of the plot and read the points of the trace.
Tools: set_view, chromatogram_image, chromatogram_data
Flags: --allow-changes for the zoom.
Check: look at the same view yourself. The assistant's reading of a picture is an interpretation, not a measurement.

9. Use a chromatogram stored in the file¶

Does sample.mzML contain stored SRM or MRM chromatograms? If so show the first one.

The assistant will: read the file's metadata, which lists stored chromatograms, and show one of them.
Tools: dataset_metadata, extract_chromatogram with native_chromatogram
Flags: --allow-changes to show it.
Check: the label of the trace matches the transition named in the metadata.

Integration¶

10. Preview an integration¶

In the XIC for m/z 258.1109 integrate the peak between 3.11 and 3.36 minutes, but do not record it. Tell me the area and the apex.

The assistant will: calculate the area and apex with a straight-line baseline and report them. Nothing is recorded.
Tools: integrate (with apply left off)
Flags: none besides --allow-root.
Check: the area is in intensity times minutes; the window's measurements list has not changed.

11. See what smoothing does¶

Extract the XIC for m/z 258.1109 twice, once with smoothing 0 and once with smoothing 5, and integrate 3.11 to 3.36 minutes in both without recording. How much does the area change?

The assistant will: extract two traces and preview two integrations.
Tools: extract_chromatogram, integrate
Flags: none besides --allow-root.
Check: the two areas differ by the percentage the assistant states. Use smoothing 0 for numbers you report.

12. Record an integration and save the table¶

Record the integration between 3.11 and 3.36 minutes and save all measurements to C:\MSData\Reports\measurements.csv.

The assistant will: apply the integration, which records it as a measurement, then write the CSV file.
Tools: integrate (with apply), measurements, export_measurements
Flags: --allow-changes, --allow-exports.
Check: the file exists, has one row per measurement, and the extraction settings are in the row. Existing files are never overwritten, so use a new file name.

13. Take back the last integration¶

Undo the last integration you recorded.

The assistant will: undo it, unless something else changed in the meantime.
Tools: undo_integration
Flags: --allow-changes.
Check: the measurement is gone from File, Integration results…. If the window changed since, the assistant reports that it cannot undo and you remove the row yourself.

Batch quantification and review¶

14. Run a method on many samples¶

Open all mzML files in C:\MSData\batch1. Read the method file C:\MSData\methods\caffeine.toml and run it on all of them. Show me the table of areas.

The assistant will: open the files, read the method, start the batch, follow its progress and summarise the results.
Tools: list_mzml_files, open_files, get_preset, start_quantification, quantification_status
Flags: --allow-changes.
Check: the Quantification and QC tab shows the batch; the method in the tab is the one you named.

15. Find the peaks that need review¶

Which results of the batch need review, and why? List the sample, the compound and the reason.

The assistant will: read every result with its confidence score and flags.
Tools: quantification_status
Flags: none besides --allow-root.
Check: compare with the colours and flags in the review table. The confidence thresholds are defaults, not calibrated values.

16. Correct a boundary¶

In sample 7, set the boundaries of caffeine to 3.15 and 3.30 minutes and tell me the new area.

The assistant will: read the trace behind the result, move the boundaries and report the recomputed area.
Tools: quantification_trace, review_quantification
Flags: --allow-changes.
Check: the peak now shows as Manual. It is not marked as reviewed; only you can press ✓ Accept peak. reset_quantification brings the automatic peak back.

17. Save the batch results¶

Export the batch results to C:\MSData\Reports\batch1-areas.csv.

The assistant will: write the table with areas, review state, settings and a snapshot of the method.
Tools: export_quantification
Flags: --allow-exports.
Check: the file has a row for every sample and compound; missing areas are empty, not zero.

Calibration, concentrations and QC¶

18. The complete workflow¶

Quantify caffeine in these 12 runs using standards 1 to 6. Show me the calibration and flag anything that needs review.

(Name the folder and the file that tells which run is which standard and at which level; the assistant cannot guess concentrations.)

The assistant will: open the runs, extract and integrate caffeine in each, build a weighted calibration from the six standards, calculate the concentrations of the others, switch the window to the calibration stage and list the flags.
Tools: open_files, analysis_capabilities, start_analysis (operations targeted_batch and quantify), analysis_status, analysis_result, show_workspace
Flags: --allow-changes so that the result appears in the window.
Check: every standard has the level you gave; the calibration range covers your unknowns; no concentration is reported outside it without a flag; missing values are empty, not zero. Look at each flagged peak yourself.

19. Explain a calibration¶

Describe the calibration curve you just built: model, weighting, range, and the accuracy of each standard.

The assistant will: read the stored calibration result.
Tools: analysis_result
Flags: none besides --allow-root.
Check: the numbers match the calibration stage of the Quantification and QC tab.

20. Evaluate batch QC¶

Evaluate the QC samples of this batch. Use a limit of 15 percent for the precision of the QC samples and tell me which rules pass, fail or are indeterminate.

The assistant will: run a QC evaluation with the limit you stated and report each rule.
Tools: start_analysis (operation evaluate_targeted_qc), analysis_status, analysis_result
Flags: --allow-changes to see the report in the window.
Check: the limit used is the one you gave, not one the assistant chose. A rule whose evidence is missing must be indeterminate, never a pass.

21. Find out why QC failed¶

QC failed for the second batch. Explain which evidence is missing or out of range, and which samples caused it.

The assistant will: read the QC report and the calls of this session.
Tools: analysis_result, session_audit
Flags: none besides --allow-root.
Check: every claim names a sample and a number you can find in the report.

Figures and exports¶

22. Save a figure¶

Save the chromatogram on screen as an SVG figure in C:\MSData\Reports\xic.svg.

The assistant will: write the figure at full resolution.
Tools: export_figure
Flags: --allow-exports.
Check: open the SVG; the axes carry units and the trace labels.

23. Save the trace as numbers¶

Save the points of the current chromatogram of sample.mzML to C:\MSData\Reports\xic.csv.

The assistant will: write retention time and intensity as a CSV file.
Tools: export_csv
Flags: --allow-exports.
Check: the first and last retention times match the trace; the file is new.

24. Save calibration and QC tables¶

Save the calibration and the QC results as files in C:\MSData\Reports.

The assistant will: write the results of the Quantification and QC tab to new files.
Tools: export_results
Flags: --allow-exports.
Check: the files exist; the call fails with a clear message if the tab holds no results.

Sessions, presets and methods¶

25. Save and continue later¶

Save my workspace as a session in C:\MSData\Reports\monday.chromascope.

The assistant will: write the session file with sources, traces, measurements and layout.
Tools: save_session
Flags: --allow-exports.
Check: the file exists. Next time, ask it to load the session; every source must still be in an allowed folder.

26. Open yesterday's session¶

Load the session C:\MSData\Reports\monday.chromascope.

The assistant will: replace the current workspace with the saved one and reopen the source files.
Tools: load_session
Flags: --allow-changes.
Check: the files and traces you expect are back. The current workspace is replaced, so save it first if you need it.

27. Build a preset for several compounds¶

Make a viewer preset for caffeine (m/z 195.0877) and theobromine (m/z 181.0720), both 10 ppm, positive mode, save it as C:\MSData\presets\xanthines.toml, then apply it to the open files.

The assistant will: write and validate the preset, then apply it to all open files.
Tools: list_presets, save_preset, apply_preset
Flags: --allow-exports, --allow-changes.
Check: open the preset in File, Viewer presets, Preset editor…: the two compounds and their masses are as you gave them.

28. Keep a method¶

Save the quantification method that is open in the Quantification tab to C:\MSData\methods\caffeine-v2.toml.

The assistant will: write the method file.
Tools: save_method
Flags: --allow-exports.
Check: the file is new and holds your compounds. Use open_method later to load it.

Projects and audit packages¶

29. Start a project¶

Create a project for this analysis at C:\MSData\Projects\caffeine-study.json and save the current state as a revision.

The assistant will: create the project from the workspace and store a first revision, then save another.
Tools: create_project, save_project
Flags: --allow-project-writes.
Check: the Project menu now shows the project name; revisions are listed and earlier ones stay available.

30. Build and verify an audit package¶

Build an audit package for the open project in C:\MSData\Reports\caffeine-package and then verify it.

The assistant will: write the package, then check every file against its checksum.
Tools: build_bundle, verify_bundle
Flags: --allow-exports.
Check: the folder contains report.html and a checksum file. If the package is refused because of open warnings, the assistant must tell you what they are; it may only continue with your say-so.

31. Read an existing project¶

Read the project C:\MSData\Projects\caffeine-study.json and summarise its revisions, datasets and saved methods.

The assistant will: read the project without changing it.
Tools: inspect_project
Flags: none besides --allow-root.
Check: the revision count equals what the Project menu shows after you open the project.

Proposals¶

A proposal is a suggested change that waits in your window until you decide. The assistant never applies it itself: it appears under Audit package, Review AI proposals, with its reason and evidence, and only your click on Approve changes a result. Say propose and don't apply in the prompt. The whole story is in Reviewing AI proposals.

32. Propose new peak boundaries¶

Check the integration of caffeine in run 3. If the boundaries look wrong, propose new ones with your reason and evidence, but don't apply them.

The assistant will: look at the batch result of that sample, read the trace behind the peak and, if the boundaries look wrong, make a proposal with the new start and end, its reason and the evidence it used. It changes nothing.
Tools: list_datasets, quantification_status, quantification_trace, review_quantification with propose, dataset_id, reason and evidence
Flags: none besides --allow-root. A proposal needs no --allow-changes; the batch must already be in the window (run it first, or open the project).
Check: the status bar says The AI proposed a change. Review opens Review AI proposals, which shows the original and the proposed start, end and area. The peak in Quantification and QC is unchanged until you click Approve.

33. Propose a review of a calibration result¶

Look at the calibration of the last run. If a sample should not be accepted, propose rejecting it with your reason and the numbers that support it, but don't apply the review.

The assistant will: read the finished calibration and, for a result it doubts, prepare a review proposal with the original and the proposed review state.
Tools: analysis_status, analysis_result, prepare_review_proposal
Flags: --allow-changes, so that the calibration is shown in Quantification and QC; you approve a proposal there against the result in the window.
Check: Review AI proposals lists the proposal with its reason. After Approve the result carries a review with both reasons; before that the table is unchanged.

34. Propose acknowledging a QC rule¶

Which QC rules failed in the last QC evaluation? For the accuracy rule, propose an acknowledgement with the reason you found, but don't apply it.

The assistant will: read the QC report, explain the failed rules and propose an acknowledgement for one of them.
Tools: analysis_status, analysis_result, prepare_review_proposal
Flags: --allow-changes, so that the QC report is shown in the window.
Check: an acknowledgement stays a note with a reason; the failed verdict does not turn into a pass, whether you approve the proposal or not.

35. Keep a proposal in the project¶

In the open project, propose rejecting sample 7 for caffeine because its qualifier ratio is outside the range. Keep the proposal in the project so that I can review it tomorrow, and don't apply it.

The assistant will: read the project, prepare the proposal and store it in the project as a suggestion. It changes no result.
Tools: inspect_project, prepare_review_proposal, queue_review_proposal
Flags: --allow-project-writes.
Check: the proposal is in Review AI proposals right away, and the project has a new revision. Close everything, open the project in the plain chromascope.exe and the proposal is in the same list.

Settings¶

36. Change the look of the window¶

Switch the window to the dark theme, interface size 115 percent and 5 decimals for m/z. Then tell me whether a newer Chromascope version is available.

The assistant will: change the preferences and read them back, including the update notice.
Tools: set_settings, get_settings
Flags: --allow-changes.
Check: the window looks as asked. The answer about a newer version comes from the program's own daily check; it may say there is none.

When a prompt does not work¶

If the assistant says a tool is refused, the message names the missing flag. See MCP troubleshooting. To get better results from your prompts, read Writing good prompts.