TOOL_CATALOG.md

Tool Catalog

The complete endpoint reference for the Predictive Maintenance MCP server: every registered MCP tool and guided prompt, grouped by category.

Total on the main branch: 42 MCP endpoints, 39 tools and 3 prompts. Tools listed under Unreleased in the CHANGELOG are not yet in a published PyPI release; their rows below say so.

Signal Lifecycle (5)

Tool	Description
load_signal	Load vibration file(s) (CSV, WAV, MAT, NPY, Parquet, raw binary .bin/.raw/.dat with declared decode metadata) with declared sampling rate and unit — returns the signal_id handle
list_signals	Browse signal files on disk (scope="disk") or loaded signals in memory (scope="memory")
get_signal_info	Signal metadata (sampling rate, duration, declared unit, source metadata)
generate_test_signal	Create a synthetic signal, auto-registered and immediately analyzable
clear_signals	Remove one signal or the whole in-memory cache

Spectral & Statistical Analysis (6)

Tool	Description
analyze_fft	Frequency spectrum with automatic peak detection
analyze_envelope	Envelope analysis for bearing fault detection (default band 500–5000 Hz)
analyze_statistics	Time-domain features (RMS, kurtosis, crest factor)
extract_features_from_signal	Segmented statistical feature extraction
compute_power_spectral_density	Power spectral density (Welch method)
compute_spectrogram_stft	Time-frequency spectrogram

Diagnostics & Health Assessment (7)

Tool	Description
assess_severity	Unified ISO 20816-3 severity assessment (signal or direct RMS reading, custom thresholds) — requires a declared signal unit, never guesses
check_bearing_faults	Unified fault-frequency matching (catalog bearing, explicit frequencies, or explicit geometry)
diagnose_vibration	Integrated evidence-based diagnosis pipeline (one call)
calculate_bearing_characteristic_frequencies	Expected fault frequencies from bearing geometry
search_bearing_catalog	Look up verified, source-traced bearing geometry
train_anomaly_model	Train novelty detection on healthy baselines
predict_anomalies	Score a signal against a trained model (bounded output)

Documentation (4)

Tool	Description
search_documentation	Semantic search over equipment manuals
read_manual_excerpt	Read pages from a manual
extract_manual_specs	Extract structured specs from PDFs
list_machine_manuals	Browse available documentation

Reporting (9)

Tool	Description
plot_signal	Interactive time-domain plot
generate_fft_report	Interactive frequency analysis report
generate_envelope_report	Envelope analysis with fault markers
generate_iso_report	Severity zone visualization
generate_diagnostic_report	Integrated diagnostic report, HTML and PDF, wording authored by the server
generate_diagnostic_report_docx	Structured Word document report
generate_pca_visualization_report	PCA anomaly projection
generate_feature_comparison_report	Cross-signal feature comparison
list_html_reports	Report management (list all or inspect one)

Prognostics (3)

Tool	Description
analyze_signal_trend	Within-recording screening: feature trend + degradation onset in one call
estimate_rul	Remaining Useful Life from repeated measurements over time (linear, exponential, Kalman) — refuses single-recording extrapolation
forecast_asset_indicator	Projection of one indicator series you supply (ISO 8601 instants with an offset, declared unit and resolution) with a naive or seasonal-naive reference method: P10, P50 and P90 per step. A projection only, never a fault finding, a failure probability or a remaining life. Unreleased: main branch only

Decision Support (1)

Tool	Description
generate_maintenance_recommendations	Maintenance recommendations from severity zone + canonical fault types

Asset Health Ledger (4)

Measurements whose companion declares a measurement object (asset, point, acquisition instant) are recorded in a local append-only ledger at load time, with a health snapshot each. These tools declare the context of a point, declare its healthy reference, read the recorded history and assess the change of a point across acquisitions. Nothing leaves the machine.

Tool	Description
declare_measurement_point	Versioned declaration of a point's context: bearing or fault orders, nominal speed, ISO 20816-3 group and support, expected unit, sensor and direction; a re-declaration reports the changed keys and how many recorded measurements need re-processing. Example: declare_measurement_point(asset_id="P-101", measurement_point_id="motor_de_h", bearing_id="6205", nominal_rpm=1800, machine_group=2, support_type="rigid", expected_signal_unit="g", expected_direction="horizontal") Unreleased: main branch only
declare_healthy_baseline	Declares which recorded measurements are the healthy reference of a point, attributed to the declarer; an empty list withdraws it. Example: declare_healthy_baseline(asset_id="P-101", measurement_point_id="motor_de_h", measurement_ids=["3f2a9c1d8e7b6a50", "b81c2d3e4f5a6978", "c0ffee1234567890"], declared_by="vibration-analyst", note="after bearing replacement") Unreleased: main branch only
get_asset_history	Index of the recorded assets (no arguments) or the history of one asset read from its ledger alone: measurements newest first with an indicator preview, point declarations, baselines, integrity. Example: get_asset_history(asset_id="P-101", max_measurements=20) Unreleased: main branch only
assess_asset_change	Change of one point against its reference (declared baseline, or the first comparable acquisitions with health not declared): band per indicator, classification with its criterion, bearing evidence over the last K acquisitions classified against the reference band on the current lineage (legacy evidence reported as recorded, unclassifiable evidence as unavailable), at most one suggested verification; reprocess=True runs one call of a re-processing sequence (at most 10 attempts, 500 file checks and 50 failures per call) and its block gives sequence_state. A call without reprocess_cursor starts a new re-processing sequence, including the call a message, a remedy or a declaration names, and a named call is an attempt, not a promise: start one only when no sequence of the point has ended in the conversation, after the user took an action a sequence listed, after a load or a declaration changed the point, once after an environment fix, or when the user asks to list the blocked measurements again. While the block gives a next_call, make exactly that call; when it gives none, sequence_state is complete or exhausted (the measurements under needs_action stay stale until their action is taken). A new sequence attempts again the failures only an attempt finds. Example: assess_asset_change(asset_id="P-101", measurement_point_id="motor_de_h", last_k=5); a next_call with a token: assess_asset_change(asset_id="P-101", measurement_point_id="motor_de_h", reprocess=True, reprocess_cursor=...) Unreleased: main branch only

Guided Workflows (3 prompts)

Prompt	Description
diagnose_bearing	Complete bearing fault diagnostic decision tree
diagnose_gear	Gear fault detection workflow
quick_diagnostic_report	Fast health screening

---

Back to the README · developer setup and extension guide: Developer's Quickstart