nih

Unbrowse for Nih (nih.gov)

Nih as tools for your agent. Unofficial: not affiliated with or endorsed by nih.gov. Unbrowse compiled these from nih.gov's own requests: each call replays the site's first-party API server-side and returns a verified result. No scraping of rendered pages, no browser on your side, no nih.gov API key.

Connect the nih.gov MCP

This server is a tool scope holding only nih.gov: its compiled tools plus the Unbrowse tools that run them, so the agent's tool list stays small and every call stays on nih.gov.

claude mcp add --transport http nih https://unbrowse.ai/mcp/nih.gov
codex mcp add nih --url https://unbrowse.ai/mcp/nih.gov && codex mcp login nih

{"mcpServers":{"nih":{"url":"https://unbrowse.ai/mcp/nih.gov"}}}

The first connect signs in to Unbrowse with OAuth (free tier, only verified results count). For scripts, send Authorization: Bearer $UNBROWSE_API_KEY (key from https://unbrowse.ai/app/keys; keep it in a secret manager, never in config you commit).

If none of the tools below are in this session's tool list, the server is not connected or not signed in: say so and stop. Do not fetch nih.gov some other way and present it as this skill's result.

Tools

`pubmed_ncbi_nlm_nih_gov__get_suggestions` — Search pubmed

Search pubmed. Inputs: query. Returns code, suggestions, term. Read-only on pubmed.ncbi.nlm.nih.gov.

query (string, required) — query (typed during “fill Search:”)
select (array) — Keep only these parts of the result: dots walk keys, [] maps over a list, a final {a,b} keeps several keys (e.g. "results[].{title,url}"). Unmatched paths come back in selectMissing.

{"query":"<query>"}

Always there too: unbrowse.run (a task on nih.gov in plain words), unbrowse.inspect, unbrowse.resume, unbrowse.scrape and unbrowse.map (any page on nih.gov), and the recorded cloud browser unbrowse.browse.open → unbrowse.browse.act → unbrowse.browse.finish. Hosts that reject dotted names show them with _ (unbrowse_run).

How to call

Pick the tool whose title matches the job and pass its inputs. Leave the site's recorded parameters out unless the user asked for something they control.
Read status. succeeded carries the verified result. input_required is not a failure: answer the open fields with unbrowse.resume {runId, answers} on the same run.
No tool fits: unbrowse.run {task}. It fails with no_capability when nothing is compiled for that yet: do it once with unbrowse.browse.open on nih.gov → unbrowse.browse.act → unbrowse.browse.finish, which compiles a new tool this server lists next time. Anything off nih.gov is refused here (not_in_scope): use the general Unbrowse server for that.
Signing in: never ask the user for a password. unbrowse.browse.act with action: "autofill" fills nih.gov's login from their Unbrowse password manager.
Report only what a succeeded result says. outcome_unknown means a change may have happened: check before retrying. Posting, buying, sending or deleting needs the user's go-ahead.

Without MCP

Every tool is also a REST call:

curl -s https://unbrowse.ai/api/v1/sites/pubmed.ncbi.nlm.nih.gov/call/pubmed_ncbi_nlm_nih_gov__get_suggestions \
  -H "authorization: Bearer $UNBROWSE_API_KEY" -H "content-type: application/json" \
  -d '{"query":"<query>"}'

OpenAPI: https://unbrowse.ai/api/v1/sites/pubmed.ncbi.nlm.nih.gov/openapi.json. Every other site: the general Unbrowse skill (https://github.com/unbrowse-ai/unbrowse) and https://unbrowse.ai/mcp.