How we search
the landscape.

Fresh searches.
Scoped additions.

The new discovery wave added 62 documentation-reviewed capability records. The current catalogue contains 4,280 core records across servers, skills, components and workflows.

Completed discovery includes 64 GitHub repository searches, 32 Smithery queries, 20 indexed-directory searches, and 216 pages from the Official MCP Registry update feed. The feed returned 21,514 latest-version update observations since September 7; this is not a count of new RWE tools or a fresh full baseline harvest.

88 candidate identities remain held for first-party identity, distinctness or component follow-up. They are outside the catalogue and separate from the historical Borderline record. The wave does not claim exhaustive coverage or a new two-round stopping result. Historical card review dates remain unchanged; new records state documentation-only review and no independent runtime testing.

Download the October refresh report (PDF) ↓ · Queries and candidate decisions (JSON) ↗

A focused core.
Discoveries preserved.

The September expansion published 6,227 cards using an overly broad interpretation of enabling infrastructure. A later relevance audit separates the core catalogue from adjacent discoveries and unresolved component questions. The website displays only the core catalogue. Archive and borderline records remain in the downloadable Excel workbook, with all existing IDs and source history preserved.

Core: a documented health-research role or a specific statistical or evidence-synthesis function. Examples include clinical/claims data, OMOP, causal inference, survival analysis, health economics, surveillance, scholarly retrieval and structured evidence review.

Discovery archive: tools whose connection depends on a possible downstream health application, including generic earthquake/climate feeds, broad public-data portals, molecular discovery and general scientific productivity. Environmental tools need an explicit health-exposure or epidemiologic scope to qualify for the core.

Borderline: a concrete unresolved component or scope question. These entries remain outside the core until resolved.

The relevance audit evaluates the recorded capabilities and limitations, with source follow-up where identified. It is not a new primary-source inspection of every project. Each card records its decision rationale and review basis. Earlier source-review dates and caveats are retained.

4,280 core · 2,008 archived · 1 borderline. One earlier source-quality hold remains separately recorded.

Download the relevance audit (PDF) ↓

Clinical Trials AI MCP.

RWE-0115 is held following a September 7 static implementation review. Its trial search, details, comparison and endpoint tools use a hardcoded simulated database. The endpoint function also infers randomization from the number of arms, which does not establish random allocation.

This component is a reference-data demonstration, not a live clinical-trial registry evidence source. Its historical ID and inclusion record remain in the workbook, with the corrected status. The other entries retain their own stated review boundaries.

Read the reviewed implementation ↗

Sources and limits.

Revised search plan

The continuation completed a revised source-specific plan. It did not complete the original broad matrix of every technology term, workflow term and source. Prospective amendments clarified homonyms and narrowed unfinished repository searches to name, description and topics; completed broad searches and partial observations remain in the audit history. This change reduces coverage of projects discoverable only through README text.

GitHub repository partition searches retained a corpus cutoff of 7 September 2026. Later run dates do not extend that lane’s repository creation-date boundary; other source lanes retain their separately recorded search rules and dates. Later Official Registry and Docker phrase screens reuse retained source bytes rather than new live inventories. Their retrieval/version dates and unavailable-source limits are recorded in the coverage report.

Sources before expansion rounds

GitHub repository discovery recorded 277 revised queries, 700 result pages and 51,103 overlapping observations. Sixteen code searches recorded 53,965 observations; their retained sharded logs do not support an independent page-by-page completeness check. Smithery recorded 214 paginated vocabulary queries, 389 pages and 4,925 unique listings. Another 48 indexed searches covered Glama, PulseMCP and MCP.so; these were ranked search results, not complete directory harvests.

The Official MCP Registry harvest recorded 28,167 latest-version records across 282 pages. Live pagination is not a transactionally frozen inventory, and automated metadata screening is not individual source review. The initial Docker checkout record counted 496 files; a later retained snapshot screen counted 569 tracked files using a different method. Neither number is a server count.

Expansion rounds and review

All 18 expansion rounds and their promoted source reviews and meaningful project links are closed. The final fixed plan used 640 fresh queries per round: 240 GitHub repository, 80 GitHub code, 80 Smithery and 240 indexed-directory queries, plus 80 phrase checks against retained Official Registry and Docker snapshots. Rounds 17 and 18 each found no new qualifying project or component after canonical reconciliation. The historical report separates 6,521 reconciled project/component units from the then-published 6,227 cards. The subsequent relevance audit determines the current core catalogue; those earlier counts are not current inclusion counts.

Search observations, primary-source reviews, canonical projects, components and public cards have different denominators. Automated scope screens, exact-copy reconciliation and individual source inspections are labeled separately. Approval of a component does not approve every sibling in its repository or collection.

Download the registry

Core, archive and borderline workbook ↗ · Core catalogue data as JSON ↗

Audit record

Historical errors, interruptions and incomplete attempts remain recorded. Successful revised searches do not retrospectively complete the original expressions. The public audit contains sanitized queries, dispositions and source references; retained raw source bodies are not republished.

Download the search coverage report (PDF) ↓ · Coverage audit data (JSON) ↗

Indexed does not mean validated.

Eligible

A public MCP server, connector, agent skill, library with an agent-facing interface, plugin, or orchestrated agent workflow that supports an identifiable RWE, HEOR, epidemiology, public-health, evidence-synthesis, or necessary enabling method. A merely conceivable downstream health application is insufficient for core inclusion.

Not eligible

Generic health assistants, incidental MCP mentions, nonfunctional demonstrations, duplicates, implausible bulk projects, or projects connected only by a broad disease or treatment word.

Deduplication and holds

Compare canonical URLs, redirects, packages, component scope and documented differences. A fork or aggregation is not automatically a new capability. Unresolved identity, mock services or material documentation conflicts may be held for further review.

Search stopping rule

Rounds 17 and 18 completed the fixed queries, terminal-page checks and meaningful source-link followups, with no new qualifying project or component in either round. The revised discovery stopping criterion was met on September 8, 2026 under the broader inclusion rules used at that time. The later relevance audit does not claim a new search or a new stopping result under its stricter boundary. This is a bounded search-process result under the recorded corpus cutoff, source access and revised query plan. It neither completes the original unamended Cartesian matrix nor establishes an exhaustive inventory of the changing public ecosystem.

Maintenance target

Weekly discovery and monthly catalogue audits are intended cadences, not evidence that a run occurred. Recorded check and review dates document completed work. New discoveries require editorial review before publication.