genomoncology/biomcp

0.8.22 — 2026-04-30

New features

• Extended WHO Prequalification drug coverage to include active pharmaceutical
  ingredients (APIs) alongside finished pharmaceutical products;
  search drug --region who --product-type api filters explicit WHO searches
  to API rows. (232)
• Added the CDC CVX/MVX vaccine identity bridge for default and EU/all
  vaccine-brand drug search, backed by local CDC bundle downloads and explicit
  refresh via biomcp cvx sync. (233)
• Added CDC WONDER VAERS as the aggregate vaccine adverse-event source for
  search adverse-event --source <faers|vaers|all>, including combined
  vaccine search, health visibility, and source inventory/docs surfaces. (235)
• Added the local GTR-backed diagnostic entity with search/get diagnostic,
  first-use bundle refresh, explicit biomcp gtr sync, and the supporting
  health/docs/spec contract. (236)
• Expanded diagnostic into a local multi-source surface with WHO Prequalified IVD
  infectious-disease products, source-aware search/get behavior, and explicit
  biomcp who-ivd sync. (237)
• Added FDA 510(k) and PMA regulatory overlays to
  get diagnostic <id> regulatory, rendering source-aware FDA device status
  when records match and a truthful empty regulatory section when they do not.
  (238)
• Added opt-in diagnostic pivots to gene and disease cards with
  get gene <symbol> diagnostics and get disease <name_or_id> diagnostics,
  backed by local GTR and WHO IVD diagnostic search. (239)
• Extended article search follow-up guidance with cross-entity keyword hints:
  JSON _meta.next_commands now includes gene and drug pivots for recognizable
  keyword patterns, while markdown See also: adds discover and
  get disease recovery paths for short disease-like phrases. (242)
• Added opt-in clinical_features disease output backed by MedlinePlus
  clinical summaries, reviewed HPO phenotype mapping, source-native evidence
  URLs, and unsupported-disease empty states via
  get disease <name> clinical_features. (252, 253, 254)
• Added offline routing via biomcp suggest <question> so worked-example
  prompts can resolve to structured _meta.workflow guidance without requiring
  a live tool call. (279)
• Added workflow-ladder sidecars and schema-backed _meta.ladder[] payloads
  so routed workflows can carry explicit ladder steps across the CLI/MCP
  contract. (282)
• Added richer CLI help examples for batch/variant-limit usage and preserved
  the parser-level skill uninstall contract through current help and MCP guard
  surfaces. (217, 224)
• Added opt-in article fulltext source-boundary coverage and PDF fallback so
  article retrieval can truthfully report source limits while recovering PDF
  text when structured fulltext is unavailable. (255, 256)
• Added DDInter-backed drug-drug interaction workflows with
  biomcp drug interactions <name>, get drug <name> interactions, local
  bundle health, source-scoped empty wording, and biomcp ddinter sync.
  (303)

Docs

• Added BioASQ-oriented how-to routing rows to skills/SKILL.md for
  gene-disease association, gene localization/protein-function, drug regulatory
  date, and variant clinical-significance workflows, steering agents to
  structured-data shortcuts before article search. (240)
• Refreshed architecture, source-integration, and current-state docs so the
  shipped architecture and CLI guidance match the post-decomposition repo
  surface. (272, 273, 274, 275)
• Repaired backtick quoting in the mustmatch study/spec contract, published
  BioASQ benchmark guidance, and made the canonical SKILL.md render/install
  contract explicit for agents and operators. (276, 280, 281)
• Aligned the public landing-copy contract with the shipped suggest and
  workflow landing bullets so make test-contracts accepts the current README
  and docs homepage feature counts. (286)
• Refreshed source, licensing, versioning, CLI decomposition, diagnostic,
  clinical-features, API-key, quality-bar, and staging-demo docs to match the
  shipped v0.8.22 surface and current local-runtime key expectations. (222,
  272, 292, 304, 306, 314, 318, 329, 332, 341, 358, 362)

Fixes

• Changed custom CLI validation failures (BioMcpError::InvalidArgument) to exit
  2, matching clap parser failures and separating bad usage from runtime
  failures. (353)
• Fixed suggest so resistance-to-drug mechanism questions anchor starter
  commands on the drug instead of filler text. (291)
• Patched rustls-webpki to the safe release line and made make check enforce
  the cargo-deny advisory gate alongside the existing license gate. (290)
• Fixed cargo-install compatibility, deterministic EMA freshness checks, OLS4
  smoke routing, WHO drug JSON envelopes, and WHO API cache-limit wording so
  install, health, and explicit WHO/API drug searches stay stable. (218, 220,
  226, 232)
• Added compact diagnostic rows and capped disease diagnostic pivots so
  gene/disease diagnostic follow-ups stay scannable and bounded. (266, 267)
• Replaced the stale GTR sample with a live-valid GTR example and added
  zero-result recovery for the local diagnostic surface. (268, 269)
• Tightened entity-aware article follow-ups with a same-session loop-breaker
  for overlapping suggestions and PubMed ESearch cleanup for bounded
  question-format filler words. (277, 278, 283)
• Improved sparse drug research-code recovery, relational discover filtering,
  trial help example/flag consistency, and CTGov intervention alias preservation
  so generated follow-up commands stay useful and shell-safe for degraded or
  ambiguous inputs. (302, 310, 313, 338, 339, 340, 342, 351, 357)
• Repaired the targeted SPEC_SMOKE_ARGS lane so it stores stable smoke
  section IDs and resolves them to current mustmatch pytest item IDs at runtime;
  the quality ratchet now checks collectability before stale line-qualified
  selectors can reach make spec-smoke. (288)
• Hardened update and machine-readable CLI contracts: biomcp update fails
  closed when checksum sidecars are missing, biomcp --json version documents
  its plain-text exception, biomcp --json list remains parseable, and
  short-literal update ratchets were replaced with structural behavior checks.
  (331, 333, 352, 355)
• Stabilized MCP stdio no-input guidance, cBioPortal study download idle/stall
  handling, disease clinical-feature specs, OLS4 disease/discover fallback IDs,
  WikiPathways parallel tests, and protein ComplexPortal specs with regression
  coverage. (326, 336, 345, 346, 350, 354, 358)
• Surfaced Semantic Scholar authentication/degradation status in article search
  and honored Retry-After during authenticated retries so throttled searches
  disclose source health and back off correctly. (364, 365, 366)

Internal

• Completed release/docs contract cleanup for the 0.8.22 line so the latest
  release notes, ticket inventory, and release metadata stay aligned. (264, 265)
• Made make check include the public Python/docs contract lane and added
  make release-gate as the named local release-blocking command, so
  release-readiness docs and validation profiles share one contract. (287)
• Reworked SPEC_SMOKE_ARGS handling around stable smoke inventory and current
  mustmatch pytest item IDs, and cleaned up .march artifact handling for
  repo submission paths. (270, 271)
• Migrated the release gate to the spec-v2 corpus, made spec-pr self-contained
  for March kickoff, restored release-gate reliability, and kept stable smoke,
  canary, cache-warm, and test-contract lanes aligned with the current shipped
  docs/spec surface. (294, 297, 298, 299, 300, 301, 307, 308, 344)
• Decomposed oversized CLI modules, absorbed residual line-cap allowlists,
  clarified the benchmark harness as internal, and added runtime-wiring/line-cap
  ratchets so structural cleanup cannot reintroduce public-surface drift. (309,
  319, 320, 321, 322, 323, 324, 325, 334, 335, 343, 347)
• Strengthened release readiness with gene-all warm-budget coverage, local gate
  uv-build fixes, leaked-artifact cleanup, binary asset attributes, CI wrapper
  stale-binary warnings, Python 3.12/Rust 1.95 CI compatibility, self-contained
  local-data spec fixtures, and final v0.8.22 readiness audits before publish.
  (247, 258, 262, 263, 285, 289, 315, 316, 317, 327, 328, 330, 348, 363, 367)
• Recorded the diagnostic-entity, HPO clinical-feature, Obsidian vault,
  biomedical news, post-v0.8.21 shipped-surface, and neural-reranking reviews
  as architecture or deferred work unless runtime wiring shipped in the bullets
  above. (230, 243, 244, 245, 246, 284)

0.8.21 — 2026-04-16

New features

• Added article date-range filtering (--date-from/--date-to and
  --year-min/--year-max aliases) to search article. (182)
• Added EMA regulatory region (--region eu) to search drug and get drug,
  backed by local EMA human-medicine feeds; data auto-downloads on first use via
  biomcp ema sync. (203)
• Added entity-aware HATEOAS suggestions (_meta.next_commands) to article
  search JSON output; gene-anchored and keyword searches return contextually
  appropriate follow-up commands. (195, 202)
• Added article search fallback suggestions when discover resolves no entities,
  so dead-end queries get recovery paths. (204)
• Surfaced EuropePMC and PubMed first_index_date per article in
  search article --json; null for sources that do not expose index dates. (196)
• Expanded trial search with drug alias union so searches match trials registered
  under any approved drug name variant. (198)
• Surfaced opt-in section hints (More: block) in gene and disease base card
  responses; get gene <symbol> and get disease <name> now show funding
  and survival opt-in hints. (209)

Fixes

• Fixed search all --counts-only --json so sections correctly omit results,
  links, and total fields per the counts-only contract. (206)
• Disambiguated FAERS 404 (drug not in post-marketing database) from "found but
  no matching events"; added ClinicalTrials.gov fallback for adverse-event
  lookups when FAERS returns 404. (197)
• Improved error messages for study download 403/404 (actionable, with
  --list recovery hint) and DisGeNET 403 (env var name and docs URL). (212)
• Added study subcommand help descriptions and a regression test so help
  completeness is enforced going forward. (210)

Docs

• Refreshed architecture docs to the post-decomposition state: source map,
  modes prose, and MCP allowlist contract. (213)
• Fixed documentation drift: README source count, CHANGELOG WHO entry, and
  root help command count. (211)
• Published two blog posts on biomcp.org: daraxonrasib six-commands workflow
  article and a follow-up that removes biomcp-python references. (200, 201)

Internal

• Added mustmatch lint for spec bash blocks so spec sections that silently
  skip due to a missing pipe are now caught. (221)
• Untracked .march build artifacts from git so worktree state stays clean. (193)
• Added Compact Approval Fields spec block for EMA and EU regulatory data
  contract assertions. (194)
• Consolidated EnvVarGuard and TempDirGuard into a central test_support
  module shared across all test crates. (199)
• Added docs-only validation profile so make docs can run without triggering
  the full build gate. (205)
• Updated tar and rustls-webpki for active RustSec advisories. (207)
• Fixed 7 make test-contracts failures from assertion drift after the
  decomposition batch. (208)

0.8.20 — 2026-03-30

Security

• Remove partial API key from health output (067)
• Harden MCP allowlist for mutating study download paths (068)

Fixes

• batch --json always includes _meta contract field (069)
• Harden GWAS variant section against live decode failures (070)
• Fix gene-to-PGx and disease-to-drug See-also routing (077)
• Fix large PMC OA full-text downloads timing out (080)

New features

• Cross-entity See-also links: drug→PGx, gene→PGx, disease→study, zero-result→discover (076)
• Quality ratchet in make check: mustmatch lint, MCP allowlist, source-registry drift (075)
• Study top-mutated helper and disease literature fallback

BioASQ product gap fixes

• Disease definitions: show full "characterized by..." text instead of truncating (080)
• Disease phenotypes: distinguish defining features from comprehensive HPO list with ranked key-feature summaries (081)
• Protein isoforms: surface UniProt isoforms in gene protein section (082)
• Drug target families: show family name (e.g., "PARP") alongside individual targets (083)
• Protein legacy names: show legacy names alongside current gene symbols (084)
• Variant legacy nomenclature: show legacy names alongside HGVS (085)
• Drug mechanism labels: improve purine-analog mechanism coverage (086)
• Protein function truncation: gene-level protein view now shows full function text (087)
• Variant-specific targets: drug targets now show variant-specific entries from CIViC (087)

Improvements

• Reworked CLI help and skill catalog for self-teaching agent navigation (072-074)
• Replaced in-shell mustmatch shim with packaged mustmatch lint entrypoint (088)

Docs

• Architecture contract docs for See-also, discover grammar, CI, post-tag proof (078)
• Documentation consistency audit across public docs surface
• Repo cleanup for layout, example paths, tracked artifacts (079)

Full changelog: https://github.com/genomoncology/biomcp/compare/v0.8.19...v0.8.20

What's New

Charts

• Added 4 new chart types: pie, heatmap, stacked bar, and scatter (now 12 total) (041, 062)
• Added waterfall chart for mutation burden visualization (062)
• Added --cols/--rows flags for terminal chart dimensions (062)
• Added --width/--height flags for SVG/PNG canvas size (062)
• Added --scale flag for PNG DPI control (062)

EMA (European Medicines Agency)

• EU drug data now auto-downloads on first use (~9 MB) — no manual setup required (057)
• Data refreshes automatically after 72 hours; biomcp ema sync for manual refresh
• search drug, get drug regulatory/safety/shortage all support --region us|eu|all

Semantic Scholar

• Cleaned up health display for unauthenticated users (055)

CI

• Moved EMA-backed specs out of PR-blocking gate (056)

Docs

• Refreshed README and docs index with marketing voice (061)
• Added 18 per-source SEO landing pages (059, 060)
• Added 3 new blog posts: agent skills, article search, cancer genomics charts

Known CI issue

• spec/03-variant.md::GWAS Supporting PMIDs failed with "error decoding response body" — this is a transient GWAS Catalog API issue, not a BioMCP bug. The test passes when the upstream API is healthy.

Full Changelog: https://github.com/genomoncology/biomcp/compare/v0.8.18...v0.8.19

0.8.18 — 2026-03-25

New sources

• Added EMA regional drug coverage backed by local EMA human-medicine feeds,
  so search drug, get drug, and get drug shortage support
  --region us|eu|all across EU regulatory status, DHPCs, referrals, PSUSAs,
  shortages, and the existing US label surface. (043, 044)
• Added the BioASQ benchmark module with benchmarks/bioasq/ingest_public.py,
  dataset and validity manifests, and the published BioASQ benchmark
  reference at docs/reference/bioasq-benchmark.md. (045, 046)

Improvements

• search drug now defaults to --region all, get drug --region is
  surfaced in help, OpenFDA label fallback prefers exact brand-name matches,
  and OpenTargets drug-indication lookups are corrected for mixed US/EU
  workflows. (043, 044)
• S2_API_KEY is now optional for Semantic Scholar-backed article
  enrichment and helpers: unauthenticated calls use the shared Semantic
  Scholar pool at 1 req/2sec, while authenticated calls keep the 1 req/sec
  quota with retry backoff. (047)