Dietary Exposure MCP v0.1.0

Dietary Exposure MCP v0.1.0

Status: stable software release; screening only

v0.1.0 is the first stable GitHub software release of Dietary Exposure MCP.
It provides a reproducible, versioned baseline for dietary-exposure screening,
assumption review, and governed evidence handoff.

This is an early 0.x release. Stable means that the package, contracts,
documentation, and release artifacts passed the declared engineering gates. It
does not mean that every scientific record has been independently approved, and
it is not a safety conclusion, scientific validation, regulator acceptance, or
authorization for a final legal, clinical, regulatory, or market-access
decision.

Release decision in plain language

The software is released for screening-only use. Scientific promotion of the
OpenFoodTox review-gated material remains open.

• The OpenFoodTox 3.0 version-7 migration contains 2,417 bulk runtime records
  and 2,417 matching provenance records. Every bulk record remains
  review_required.
• The 16-record high-impact review is complete and temporally current. Its
  canonical content SHA-256 is
  0feb8e3e4f9852c2d102375dd89d814ed08407a602d699882cf48bdd7f3c8c90.
• The first independent scientific review returned not approved. Its four
  blocking findings were remediated, but positive independent toxicologist or
  dietary-risk-assessor signoff has not yet been recorded.

Publishing v0.1.0 does not change those scientific states. Affected records
remain screening and review material until a qualified reviewer explicitly
approves a later versioned packet.

Highlights

• 49 typed, read-only MCP tools with closed-world annotations
• deterministic acute and chronic dietary-intake calculations
• bounded, survey-summary, cohort-bootstrap, and uncertainty-support workflows
• 2,462 governed reference-value records across curated and review-gated layers
• reference-value, legal-limit, method, occurrence, and consumption registries
• contaminant, metals, trade-risk, interoperability, and scientific follow-up
  review artifacts
• PBPK-ready oral dose handoffs without PBPK execution
• 98 published JSON schemas and 90 generated examples
• version-pinned defaults, manifests, validation dossiers, and source provenance

OpenFoodTox 3.0 scientific corrections

The migration preserves raw source encoding while keeping canonical scientific
interpretation explicit:

• Acetamiprid ARfD is represented as 0.005 mg/kg bw; the workbook's daily
  encoding remains visible only in raw provenance.
• Glyphosate ARfD is represented as 1.5 mg/kg bw and classified as a
  primary-source unit correction rather than an exact unit match.
• Imidacloprid retains the 0.08 mg/kg bw MRL-assessment context while recording
  the separate 2013 recommendation of 0.06 mg/kg bw; neither is silently
  flattened into an unconditional supersession claim.
• All non-null workbook field paths must resolve to real headers. The 564
  lower-bound ARfD records that previously named a nonexistent qualifier column
  now carry a null field path.

See the
independent-review remediation record,
scientific source review, and
migration ADR.

Verification

The complete release verifier passes from the release commit:

• 360 tests
• 455 generated validation cases across 35 suites
• pinned-workbook OpenFoodTox 3.0 migration verification
• MCP conformance, schema-drift, and scientific-invariant gates
• Ruff, Bandit, dependency auditing, and full-history secret scanning
• reproducible wheel and source distribution builds
• Twine metadata checks and clean-environment wheel installation
• CycloneDX SBOM and SHA-256 release checksums

GitHub's supply-chain, secret-scan, Semgrep, MCP-conformance, schema-drift, and
scientific-invariant checks also pass.

Install and verify

Download all release assets into one empty directory:

gh release download v0.1.0 --repo ToxMCP/dietary-exposure-mcp
shasum -a 256 -c SHA256SUMS
uv tool install ./dietary_mcp-0.1.0-py3-none-any.whl
dietary-mcp

The attached distribution reports Python package version 0.1.0. It is
distributed through this GitHub release and is not published to PyPI.

Known limitations

• Positive independent scientific signoff is not complete.
• Imported OpenFoodTox bulk records remain review_required; they are not
  curated or regulator-approved merely because the software is released.
• Reference values, MRLs, legal limits, and guidance may change after the pinned
  snapshots; decision-relevant values require current primary-source checking.
• Results depend on the quality and applicability of residue, consumption,
  body-weight, processing, mapping, and censoring assumptions.
• The bounded survey and uncertainty workflows are not a universal population
  model.
• The service is not a final regulatory decision engine, PBPK execution engine,
  proprietary PRIMo/DEEM/DietEx engine, or submission portal.
• Supported v0.1.0 operation is local stdio. A hosted or public HTTP deployment
  requires a separate security review and authenticated gateway.
• WHO-derived GEMS/Food seed-profile redistribution requires separate attention
  for the intended audience and commercial use.
• v0.1.0 remains on the stable MCP v1 SDK line; MCP v2 beta migration remains
  opt-in and isolated.

The full explanation and safe-use checklist are in
Limitations and Intended Use.

Reporting and attribution

Use the repository's scientific-correction form for source, unit, population,
or interpretation concerns. Use GitHub private vulnerability reporting for
security issues. Do not place credentials, personal data, confidential studies,
or unpublished dossiers in public issues.

Third-party attribution and data-rights boundaries are documented in
THIRD_PARTY_NOTICES.md.
Original authority outputs remain the scientific and legal sources of record.