Amass Connector

Search 60M+ biomedical records — papers, trials, drugs, genes, FDA/EMA approvals, patents — from an AI agent.

Add to Agentman Studio

Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.

Paste it into any MCP client. Setup docs

Resources

What the Amass Connector connector does

The Amass MCP server connects a unified biomedical index — peer-reviewed papers, clinical trials, drugs, genes, FDA and EMA approvals, and patents, cross-linked to each other — to Claude, ChatGPT and any MCP-compatible agent. All 13 tools are read-only searches and record lookups. Sign-in is OAuth against an existing Amass account, which Amass includes on every plan.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Amass Connector tools (13)

  • get_amass_biomedcore_record
  • get_amass_drugcore_record
  • get_amass_genecore_record
  • get_amass_patentcore_record
  • get_amass_regulatorycore_document_section
  • get_amass_regulatorycore_record
  • get_amass_trialcore_record
  • search_amass_biomedcore_records
  • search_amass_drugcore_records
  • search_amass_genecore_records
  • search_amass_patentcore_records
  • search_amass_regulatorycore_records
  • search_amass_trialcore_records

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • Amass publishes its own recall problem, and it is the most important limit here. Amass's benchmark post states that recall against a hand-curated ground truth "is still in the 17–21% range". Its four-server MCP comparison scores Amass at 33.6 percent average F1 — the best of the four tested, against PubMed at 21.7 percent, Consensus at 26.5 percent and Paperclip at 22.7 percent, but still a minority of the known-relevant literature. An agent using this connector is reasoning from a partial evidence set, and Amass says so plainly.
  • No application scopes exist. The RFC 9728 descriptor and the authorization server both advertise only openid, profile, email and offline_access. Nothing in the grant separates one core from another. With an all-read surface the consequence is limited, but a future write tool would arrive inside the same undifferentiated grant.
  • No result pagination. Amass documents that its API has "no cursor, no offset" and no sort parameter, with limit capped at 300 per call and 200 for PatentCore. A question needing more than 300 hits must be narrowed with filters rather than paged, which bounds how exhaustive any single landscape scan can be.
  • PatentCore is a preview. Amass states its schema "may still change while in preview", and that search collapses each patent family to one publication — so a family-level view requires a follow-up rather than appearing in search results.
  • Cross-links are unevenly populated. Amass documents that a drug record's referencesBiomedCore links are "currently sparse (single digits, sometimes empty)" and instructs readers to treat an empty list as "no links recorded, not no evidence". The same caution applies to gene target-intelligence fields, which Amass says are null when Open Targets holds no data.
  • The journal-quality filter is narrower than it looks. Amass documents minJournalQualityJufo as the Finnish JuFo ranking, and warns that "many legitimate non-Finnish journals are null" and that filtering excludes every unranked record. Using it as a general quality gate silently drops most of the world's journals.
  • We could not read tool schemas or safety annotations. The endpoint returned 401 to an anonymous request, so no parameter-level detail on this page comes from the server itself; it comes from Amass's published API reference for the same six cores.
  • Rate limits on the MCP path are undocumented. Amass publishes 60 requests per 60-second window for the Platform API, but publishes no figure for MCP, where limits are described only as fair use. Treat the MCP rate ceiling as unknown.
  • Amass's own page understates its listing tier. Amass's MCP page describes itself as listed in Claude's connector directory "as a community connector". Anthropic's directory listed it as community on 2026-08-16 and as partner on 2026-08-23; the vendor's copy has not caught up. The badge on this page reflects Anthropic's current listing.
  • Certifications are in progress, not held. Amass states its SOC 2 and ISO 27001 controls are "fully mapped" with "formal certification in progress", estimated at roughly six and twelve months respectively. Amass offers right-to-audit provisions and questionnaires under NDA in the interim. If your procurement requires a certificate today, there is not one.
  • One internal inconsistency on the vendor's pricing page. It says both "All 8 data cores" and "All 6 data cores" on the same page. Six cores are named everywhere else, including the connector description and the API reference, so six is the figure this page uses.

Frequently asked questions

Is the Amass MCP server the same as the OWASP Amass subdomain tool?

No. They share a name and nothing else. This connector is built by Amass technologies ApS of Copenhagen and searches biomedical literature, clinical trials, drugs, genes, regulatory approvals and patents. OWASP Amass is an open-source network reconnaissance tool for subdomain enumeration. Different companies, different domains, no shared code or endpoint.

Can the Amass connector write or change anything?

No. All 13 tools are read-only lookups. Seven names begin with get and six begin with search, and every one of them retrieves records from an Amass index. No tool creates, updates, deletes, uploads, exports or sends anything, and Amass's own skill files describe the same 13 names as search and lookup operations only.

What OAuth permissions does the Amass MCP server request?

Only identity claims. The server's RFC 9728 descriptor advertised openid, profile, email and offline_access on 2026-08-23, and its authorization server advertises the same four. None of them names an object or an action, so the consent screen offers no way to narrow which cores an agent may read. The read-only tool surface is what limits it, not the grant.

Does using the Amass MCP server cost money per query?

Not on the MCP path as Amass documents it. Amass states MCP is included on every plan including Free, governed by a fair-use policy where heavy users get a usage conversation rather than a hard cut-off. The per-call prices Amass publishes — 0.05 US dollars per 20 search results — apply to the separate Platform API product, which is billed independently.

Which tools does the Amass MCP server provide?

Thirteen, one search and one get per data core, plus one extra. The six cores are BioMedCore for papers, TrialCore for trials, DrugCore for drugs, RegulatoryCore for FDA and EMA approvals, GeneCore for genes and PatentCore for patents. RegulatoryCore adds a thirteenth tool that fetches the full text of one parsed label or review section.

How accurate is Amass at finding the right papers?

Amass publishes its own numbers and they are modest in absolute terms. Its 10-question benchmark scored Amass at 33.6 percent average F1 against PubMed at 21.7 percent, and a separate post states recall against a hand-curated ground truth sits in the 17 to 21 percent range. Treat retrieved evidence as a starting set, never as a complete one.

What do I need before connecting Amass to Claude?

An Amass account and a paid Claude plan. Amass states the connector works on Claude Pro, Max, Team and Enterprise, and that setup is a one-time sign-in with an existing Amass account with nothing else to configure. A free Amass account is enough, because Amass includes MCP access on every plan including Free.

Does the Amass connector expose patient data or personal information?

It reads published records, not patient records. The six cores draw on PubMed, ClinicalTrials.gov and the WHO ICTRP, ChEMBL, FDA and EMA filings, and patent registries — all public. Author names, ORCIDs and institutional affiliations do come back on publication records, and inventor and assignee names on patents, because those are printed on the documents themselves.

Sources

More Healthcare connectors

  • AdisInsight

    Query Springer Nature's drug-pipeline and trial database from an AI agent.

  • AdvancedMD

    Look up patients, find open slots, book appointments, and work denials and remittances in AdvancedMD from your AI agent.

  • AllTrails

    Search AllTrails' trails by place, map bounds or name and pull details and a 7-day trailhead forecast from your AI…

Browse all MCP connectorsClaude and ChatGPT setup guideBrowse AI skills

Product names and logos belong to their owners. Agentman is not affiliated with a connector's maker unless the page says the connector is built and run by Agentman.

Use it in an agent

Put Amass Connector to work.

Connect Amass Connector once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio
Call (650) 285-1019Our AI receptionist answers.