The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Ukri Gtr listing page.
UKRI Gateway to Research MCP — every research grant awarded by the UK's seven research councils (EPSRC, MRC, BBSRC, NERC, ESRC, AHRC, STFC) plus Innovate UK: who was funded, for what, how much, and when. Keyless.
Part of Pipeworx — an MCP gateway connecting AI agents to 1476+ live data sources.
gtr_search_projects(query, funder?, status?, search_field?, limit?, page?) — free-text search over UK grants; returns grant reference, title, status, funding council, grant category, lead department and funding period.gtr_get_project(project) — one project by GtR GUID or by human grant reference (EP/T022159/1, 10045762). Follows the FUND and LEAD_ORG links to return the award value in pounds, the lead organisation and its address, the principal investigators, and the full abstract / technical abstract / potential-impact text.gtr_search_organisations(name, limit?, page?) — find funded UK institutions by name; returns organisation id, name, postcode and region.gtr_organisation_projects(organisation, funder?, status?, limit?, page?) — one institution's funding portfolio, by organisation id or by name.gtr_search_publications(query, search_field?, limit?, page?) — journal articles and papers grant holders reported as outcomes; returns DOI, PubMed id, journal, date and the originating project id.None. Keyless and unmetered — no key, account or registration.
Two headers are mandatory and are set by the pack: Accept: application/vnd.rcuk.gtr.json-v7 (the vendored media type; a plain Accept: application/json gets you the wrong representation) and a normal browser User-Agent.
https://gtr.ukri.org/gtr/apihttps://gtr.ukri.org/gtr/api/projects?q=cancer&s=10&p=1https://gtr.ukri.org/gtr/api/projects/{guid}https://gtr.ukri.org/gtr/api/funds/{guid}https://gtr.ukri.org/gtr/api/organisations?q=imperial+college&s=10&f=org.nhttps://gtr.ukri.org/gtr/api/organisations/{guid}/projectshttps://gtr.ukri.org/gtr/api/outcomes/publications?q=cancer&s=10The page-size parameter s is rejected outside 10–100 — s=1 returns HTTP 400 Page size cannot be less than 10, s=200 returns Page size cannot be greater than 100, so the pack clamps every request into that band and slices the result down to the caller's limit. The award value is not on the project object: money lives on the fund reached through links.link[] where rel === 'FUND', and gtr_get_project follows that link (plus LEAD_ORG and PI_PER) so an agent asking what a grant was worth gets an answer in one call — a fund's amount can legitimately be 0. The project's own start and end fields are null on essentially every record (0 of 100 populated in a sample), so the pack takes the funding period from the FUND link's start/end instead; both are epoch milliseconds and are emitted as ISO dates. The collection key in the response envelope changes per endpoint — project, organisation, person, fund, publication — and the links.link[] array is large and useless to a model, so it is stripped from all output and used only internally.
The nastiest trap: /projects/{guid} never 404s. Given an unknown GUID it matches on GUID segments and returns HTTP 200 with a completely unrelated project (00000000-0000-0000-0000-000000000000 returns a BBSRC studentship whose id merely contains 0000), so gtr_get_project compares the returned id against the requested one and reports found: false rather than answering about the wrong grant. /organisations/{guid} does error correctly. A search that matches nothing also 404s, with Page requested exceeds maximum available: 0, which the pack converts into an empty result rather than an exception.
GtR has no server-side filter for funder or status. Its search does honour a Lucene-ish AND across a multi-field query, so pairing the text field with pro.lf narrows to roughly 95% pure upstream; the pack does that and then matches leadFunder and status exactly client-side, walking up to 5 upstream pages of 100 to fill the requested limit. Matches in a big portfolio cluster unevenly across pages, so a found: false from a filtered call returns the page to resume from rather than implying the grants do not exist. Restricting a search to one field uses f= — valid project fields are pro.t (title), pro.a (abstract), pro.gr (grant reference), pro.lf (lead funder); anything else returns HTTP 400 naming the invalid field.
Organisation search is unusable without f=org.n: the default multi-field search for imperial college returns Jilin University and NYU, while the name-scoped search returns the Imperial records. GtR then fragments each institution across dozens of duplicate records — "University of Cambridge" has 40+, holding 6,949 / 1,547 / 305 / 163 grants respectively, and the largest is not the highest-ranked. So gtr_organisation_projects resolves a name by probing up to 20 candidates for their grant count, keeps the fullest, and returns the runners-up as alternatives with their counts. Address fields carry the literal string "Unknown" rather than null, which the pack normalises. Finally, /organisations/{id}/projects counts a grant when the institution leads or merely participates, so a row's lead_organisation_id is often a different body. Titles and abstracts arrive HTML-escaped (Simulation & Analytics) and are decoded.
Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):
tools/list at https://gateway.pipeworx.io/ukri-gtr/mcp returns the tools in the table
above plus the shared Pipeworx meta-tools — ask_pipeworx,
discover_tools, search_within, remember/recall and the rest of the
gateway-wide set. So the tool count you see is larger than this table: a
single-pack endpoint currently lists roughly 30 shared tools alongside the
pack's own. The connection's initialize response states its exact scope, and
is the authoritative answer for a given day.
This is deliberate, not multiplexing by accident. The meta-tools are what let a
scoped connection answer a question this pack does not cover — via
ask_pipeworx, which routes across the whole catalog — without you adding a
second MCP server. There is currently no way to mount a pack endpoint without
them; if the extra schemas cost you more context than the routing is worth,
connect to the full gateway once rather than to several pack endpoints.
Or connect to the full Pipeworx gateway to get every pack's tools listed directly, instead of just this one's:
Both URLs reach the same gateway and the same 1476+ data sources. The
only difference is which pack's tools are listed directly; ask_pipeworx
reaches all of them from either one.
Instead of calling tools directly, you can ask questions in plain English — this works on the pack endpoint above as well as on the full gateway:
The gateway picks the right tool and fills the arguments automatically.
MIT