The Model Context Protocol (MCP) server referenced as tech.amass/amass is described as enabling linked life-science search covering 40M+ papers and 1.2M+ trials, including information on drugs, genes, FDA/EMA approvals, and patents. The provided metadata includes topics related to bioinformatics and model-context-protocol.
๐ ๏ธ Key Features
Linked life-science search
Coverage includes papers (40M+) and trials (1.2M+)
Linked entities: drugs, genes
Includes FDA/EMA approvals and patents
๐ Use Cases
Life-science discovery across papers and trials
Investigations involving drugs and genes
Reference and linkage to FDA/EMA approval and patent information
โก Developer Benefits
Includes MCP and MCP-server context via provided topics
Topic alignment for developers working on life-sciences and clinical-trials data
โ ๏ธ Limitations
No toolCount, specific tool names, or operational details are provided in the source data.
Amass is a remote MCP server for life-science
research. It exposes one linked corpus โ publications, clinical trials, drugs,
genes, regulatory approvals and patents โ to Claude, ChatGPT, VS Code, Cursor and
any other MCP client.
This repository holds the MCP Registry
listing (server.json). The server itself is a hosted service, so there is
nothing here to install or run.
Connect
Amass speaks streamable HTTP with OAuth 2.1. Most clients need only the URL:
code
https://mcp.amass.tech/mcp
You will be redirected to sign in with your Amass account; new accounts are
provisioned automatically with trial credits. Clients that cannot complete an
OAuth flow can instead send an Amass API key (amass_โฆ) as a bearer token.
Or install it from the registry by name: tech.amass/amass.
The server supports every MCP protocol revision from 2024-10-07 through
2026-07-28. Clients on 2026-07-28 are served per request, with no initialize
handshake or session; the tools are identical on every revision.
What you get
Six Cores, cross-linked so you can move between them in a single query โ
publications โ trials, trials โ drugs, drugs โ genes, drugs โ approvals, and
patents โ drugs/publications.
Core
Contents
BioMedCore
43M+ biomedical publications (PubMed/PMC-derived) with abstracts, authors, journal metadata, citations, MeSH terms and optional fulltext, plus conference abstracts from society meetings
TrialCore
1.2M+ clinical trials (ClinicalTrials.gov + WHO ICTRP non-US registries) with phase, status, sponsor and collaborators, conditions, interventions and arms, enrollment, registered endpoints and posted results
DrugCore
22K+ harmonized drugs/molecules (ChEMBL-anchored) with modality, clinical stage, chemical structure, synonyms and trade names
GeneCore
43K+ harmonized genes (Ensembl-anchored) with biotype, RefSeq summary, tractability/safety intelligence and UniProt protein annotations
RegulatoryCore
Unified FDA + EMA authorizations with approval status, orphan designations, and parsed source documents (labels, SmPCs, reviews, EPARs)
PatentCore(preview)
Life-science patents (USPTO/EPO/WIPO subset) with full-text search, CPC/IPC classifications, assignees/inventors and patent families
Each Core has a search_amass_<core>_records tool for discovery and a
get_amass_<core>_record tool for fetching a known identifier. RegulatoryCore
adds get_amass_regulatorycore_document_section for full source-document text,
and send_feedback reports data issues straight to the Amass data team.
Every search tool takes a limit (1โ50, default 10). Every search except
PatentCore also takes minLastUpdateDate and minCreateDate, which filter on
when Amass last updated a record or first ingested it. Use them for "what
changed since I last looked" questions. Publication, patent, trial start and
trial completion dates can be bounded on both ends, and BioMedCore search can be
narrowed to given journals (journalIssns) or publication types
(publicationTypes, e.g. Clinical Trial or Conference). Every record
carries a url that links to its source.
Everything below is the release runbook for this listing. It is not needed to
use the server.
server.json is the source of truth. The registry authenticates by DNS: a TXT
record on amass.tech carries the public half of an Ed25519 key, and CI signs
with the private half. That is why the listing is namespaced tech.amass/โฆ
rather than under a GitHub org.
Publishing a new version
Each ## vN entry in the MCP changelog
becomes listing version N.0.0, but only once the server is actually deployed.
Check that the new parameters appear in the live tool schemas before bumping.
On a release-N branch, bump version in server.json. Update the README's
"What you get" for anything user-facing in the changelog entry, and update
description if a headline count moved.
Open a PR โ CI validates the file, checks the version is not already
published, and confirms https://mcp.amass.tech/mcp is up.
Merge, then tag:
bash
git tag vX.Y.Z && git push origin vX.Y.Z # must match "version" in server.json
A second admin approves the mcp-registry deployment on the tag's Actions
run. Publishing waits there until they do.
Create the GitHub Release and point back to the changelog:
bash
gh release create vX.Y.Z --verify-tag --latest \
--title "vX.Y.Z โ aligned with MCP server vX" --notes "โฆ"
Delete the local release-N branch. GitHub drops the remote one on merge.
The tag must match version in server.json or the workflow fails.
The listing version tracks the major version of the deployed MCP server.
One-time DNS auth setup
bash
./scripts/setup-dns-auth.sh
Generates the keypair and prints the Cloudflare TXT record and the
gh secret set command for MCP_PRIVATE_KEY. Follow its four steps, then
delete the local key.pem.