MoTeC MCP 0.3.0

MoTeC MCP 0.3.0

This minor release adds a public, bounded telemetry-analysis foundation and the
portable guidance another Codex installation needs to use it safely.

Highlights

• Adds four strict, read-only MCP tools for configured-root CSV logs:
  motec_log_summary, motec_log_channels, motec_log_query, and
  motec_log_quality_report.
• Captures one exact source snapshot, hashes and revalidates the opened file, and
  processes selected numeric/empty/invalid states in the same bounded streaming
  pass. Default summaries retain no sample projection.
• Supports exact channel/unit discovery, bounded cursor paging, inclusive row/time
  filters, deterministic uniform downsampling, exact numeric aggregates, and
  aggregate-only quality observations for missing/invalid values, configured-time
  behaviour, clipping candidates, and constant/stale runs.
• Binds cursors to a process-local HMAC plus the exact source and normalized CSV
  interpretation. Changing the digest, parser/query version, delimiter, layout,
  channel identities, time column, filters, order, or budgets invalidates a cursor.
• Separates source completeness from response-page completeness and reports every
  threshold as explicit, default, or fixed provenance rather than presenting an
  unexplained diagnosis.
• Adds portable i2/RaceStudio logging guidance, official-source metadata, local
  installed-help discovery, fresh-Codex routing, synthetic examples, and exact
  questions for missing source/session/export evidence.
• Adds the MIT licence with copyright © 2026 Herbertmt978 and refreshes the
  professional README, plugin metadata, tags, and installation hand-off.
• Makes Windows prerequisite probing tolerant of ordinary cold npm startup and
  runs long PowerShell automation from short-lived script files instead of an
  oversized command line.

Safety and evidence boundaries

• CSV analysis is vendor-neutral. No MoTeC .ld or AiM
  .xrk/.xrz/.drk/.gpk decoder, dialect, export, lap, beacon, resampling, or
  native-parity claim is made.
• A vendor-specific claim still requires a private native original, a paired
  same-session CSV, exact installed application/build and activation/licence
  evidence, and at least three human- or UI-confirmed facts.
• Query output contains only selected numeric, empty, or invalid states. It never
  returns raw CSV rows, raw text cells, source bytes, requested paths, or physical
  paths. Private logs and generated evidence stay in authorised local roots and
  must not be committed to Git or included in the release.
• Quality findings are observations with evidence counts and threshold
  provenance, not proof of a sensor, wiring, CAN, mechanical, calibration, or
  vehicle fault. The workflow explicitly stops tuning around unresolved faults.
• The new tools do not launch i2/RaceStudio, retrieve or erase device data, change
  a workspace, write cache state, communicate with hardware, or authorize a live
  action.
• PDM section capture now skips full-smoke only through a structured, path-free
  initial-preflight result. A live selection verifies caret/viewport ownership and
  an unchanged Windows input epoch, but Win32 TreeView messages are not atomic
  with human input; do not use mouse or keyboard while that capture runs.

Compatibility and installation

The four tools and documentation are additive; existing MCP tool contracts remain
available. A fresh Windows install requires PowerShell 5.1 or 7, Node.js 20+ with
npm, Git (with GitHub CLI optional), Codex CLI/Desktop for registration and use,
the repository, a main MoTeC data root, and any additional roots the user chooses
to authorize.
MoTeC/AiM applications and installed help are optional for reduced offline use and
required only for their corresponding full/live capabilities. Logs and CSV exports
are not installation prerequisites. Installed i2 readiness proves declared files
are present only; it does not prove edition, activation, API entitlement,
workspace load, or native export capability.

Follow setup from GitHub, then run the checks in the
release checklist. Native-log parity remains a future
evidence gate and is not implied by a green build, smoke test, or release.