Elicit
by Elicit
Search 138M+ academic papers and ClinicalTrials.gov records from your AI assistant, and launch AI-written literature reviews. Reports spend Elicit Credits and take 5-15 minutes. Elicit's own docs list 8 tools where Anthropic's directory lists 5.
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.
Connect Elicit via MCP
https://elicit.com/api/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Elicit Tools & Capabilities (5)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- Elicit can hallucinate, and says so. Its help centre states Elicit may state what seems likely rather than what the papers say, and that it can "miss the nuance of a paper or misunderstand what a number refers to" (Elicit help centre, fetched 2026-08-19).
- Elicit does not judge study quality. It states it cannot yet evaluate whether one paper is more trustworthy than another, and summarises the findings of a bad study the same way it summarises a good one. A journal-quartile filter is not a quality judgement.
- Elicit is "only as good as the papers underlying it." Its stated limitations page says so directly, and adds that the list given there is not comprehensive.
- Coverage skews to empirical research. Elicit describes itself as working better on empirical work — randomised trials in the social sciences and biomedicine — than across all domains equally.
- The write tools spend money and the cheap one has no preflight.
create_reportcommits credits the moment it is called; onlycreate_systematic_reviewoffersdryRunpricing. - The credit estimate is not a quote. Elicit describes it as median-anchored, warns that broad searches can cost more than shown, and notes API-launched reviews always screen in strict mode.
- Running out of quota pauses rather than fails. A report or review enters
pausedForInsufficientQuotamid-run and stays there until explicitly resumed;resume_sessionfails while the account is still over its limit. - Pro plan or above is required. Elicit documents Basic and Plus as having no API access, and returns HTTP 403 telling those tiers to upgrade.
- Result caps are plan-gated at 300 per request on Pro, 500 on Scale and 10,000 on Enterprise, regardless of what
maxResultsis set to. - Rate limit is 100 requests per minute per IP, applied across all endpoints and all plans; exceeding it returns HTTP 429 and blocks the IP for 5 minutes.
- Reports can be made public.
create_reportandcreate_systematic_reviewboth take anisPublicflag; Elicit documents it as defaulting to false, which means the default is safe and the unsafe value is one argument away. - We could not verify tool behaviour ourselves. The endpoint returned HTTP 401 to an anonymous handshake on 2026-08-19, so no schemas and no safety annotations were readable. Tool names come from Anthropic's directory and Elicit's README; the read-only markings are Elicit's.
- Prompts are unconfirmed. Anthropic's snapshot lists no prompt names, and the gated endpoint blocked our own
prompts/list, so we cannot say whether the server serves any.
Frequently asked questions
Does create_report cost money, and can I see the price first?
It spends Elicit Credits, and for reports there is no preflight. Elicit's API documentation gives a `dryRun` cost preview only to systematic reviews, which returns an estimated credit total and a per-component breakdown without charging. The report endpoint has no equivalent parameter, so an agent calling `create_report` commits the spend at the moment it calls. Check your balance first.
How long does an Elicit report take, and is the call asynchronous?
Reports typically take 5 to 15 minutes and the call is asynchronous. Elicit's documentation says `create_report` returns immediately with a session ID and a URL, then you poll `get_report` until the status reads completed or failed. Elicit recommends polling every 30 to 60 seconds. The full report body is omitted from poll responses unless you ask for it.
What corpus does Elicit search, and where do the clinical trials come from?
Two different corpora. Elicit documents semantic search across more than 138 million academic papers, with a `pubmed` option that restricts results to PubMed-indexed literature. `search_trials` is separate: Elicit states its trial records come from ClinicalTrials.gov, and it takes trial-specific filters for phase, recruitment status and whether results have been posted.
Does the Elicit MCP server list every tool Elicit documents?
No. Anthropic's 2026-08-16 snapshot names five tools; Elicit's own MCP README documents eight. The four the snapshot omits are `create_systematic_review`, `get_systematic_review`, `list_sessions` and `resume_session` — two of which are writes. The snapshot also names `list_reports`, which does not appear in Elicit's README under that name.
Can Elicit hallucinate, and should I check the papers myself?
Yes, and yes. Elicit's help centre states it can experience hallucination, stating what seems likely rather than what a paper says, and that it can miss a paper's nuance or misunderstand what a number refers to. Elicit also says it cannot judge whether one paper is more trustworthy than another, summarising a weak study exactly as it summarises a strong one.
What plan do I need for the Elicit MCP server?
Pro or above. Elicit's MCP README lists a Pro-plan account as a requirement, and its API documentation returns HTTP 403 telling lower tiers to upgrade. Plan tier also caps results per search request: Elicit documents 300 for Pro, 500 for Scale and 10,000 for Enterprise, with Basic and Plus getting no API access at all.
What happens if my Elicit account runs out of quota mid-report?
The run pauses rather than failing. Elicit documents a `pausedForInsufficientQuota` status that a report or review enters when the account exceeds its usage limit mid-run, and it stays paused until someone explicitly resumes it. The `resume_session` tool restarts it, and Elicit says that call fails while the account is still over its limit.
Sources
- Elicit MCP README — https://github.com/elicit/api-examples/blob/main/integrations/mcp/README.md (fetched 2026-08-19 as raw Markdown via
raw.githubusercontent.com; the only file underintegrations/mcp/in the repository) · retrieved 2026-08-19 - Elicit API reference, machine-readable index — https://docs.elicit.com/llms.txt (fetched 2026-08-19; OpenAPI 3.1.0, API version 2.0.0 — the source for credits,
dryRun, plan caps and rate limits) · retrieved 2026-08-19 - Elicit help centre, stated limitations — https://support.elicit.com/en/articles/549569 (fetched 2026-08-19) · retrieved 2026-08-19
- Elicit help centre, hallucination and reliability — https://support.elicit.com/en/articles/552897 (fetched 2026-08-19) · retrieved 2026-08-19
docs.elicit.com/robots.txtisAllow: /with a sitemap, and carries noContent-Signaldirectives (fetched 2026-08-19).elicit.com/robots.txtreturns the site's 404 page, andelicit.com/sitemap.xmlreturns HTTP 404 · retrieved 2026-08-19- Live RFC 9728 protected-resource descriptor — https://elicit.com/.well-known/oauth-protected-resource/api/mcp and https://elicit.com/api/mcp/.well-known/oauth-protected-resource (both fetched 2026-08-19, byte-identical; the root form returns 404) · retrieved 2026-08-19
- Live RFC 8414 check — https://elicit.com/.well-known/oauth-authorization-server returned 404 on 2026-08-19, as did its path-insert variant; Elicit's README names
elicit.com/api/auth/.well-known/oauth-authorization-serverinstead · retrieved 2026-08-19 - Live anonymous probe — 2026-08-19:
POSTto the endpoint returns HTTP 401 withwww-authenticate: Bearer realm="Elicit MCP", resource_metadata="…", scope="elicit.mcp", error="missing_authorization"· retrieved 2026-08-19 - Anthropic connector directory — https://claude.ai/directory/1287875c-308f-4a61-9ebf-4a0201ef214f (snapshot 2026-08-16; permissions label "Read and write", five tool names, no prompt names) · retrieved 2026-08-16
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Healthcare
- Developer
- Elicit
- Tools
- 5
- Domain
- elicit.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.