browse_research_records
ChatGPTExplore a collection returned by list_data_collections in bounded pages. Organizations: name query/EIN/state. Funding: funder or recipient name/EIN/recipient state. Papers: title/abstract/population query, exact paper record_id, or relevance filter (direct/adjacent/unrelated/uncertain); includes versioned reviews and text basis, alongside legacy classifications. Programs: name/description/intervention query or exact program record_id, with source paper links. Outcomes: metric/paper-title query or record_id meaning source paper_id. Reports: organization/title query or EIN. Unsupported filters are rejected. No arbitrary SQL, bulk export, private documents, or write access. Use compare_org_financials for filings. Present concise text; use a native table only when it helps compare records or the user requests it.
compare_org_financials
ChatGPTRead up to three filing tax years for one to ten selected EINs, with individual source links, contribution/revenue metrics, duplicate and conflict handling, and explicit unknowns. Non-government and external contribution ratios are calculated only from known compatible inputs. This does not select comparable peers, remove campaign/transfer effects, or set a fundraising target. Default to a concise text answer; use native charts/tables when a multi-year trend or selected-peer comparison benefits from them, or when requested. No automatic UI panel.
expand_issue_funder
ChatGPTFor a selected funder EIN, state, and grant-purpose terms, show discrete filing rows for other recipient organizations. Purpose mentions can be mixed, and recipient state is an address rather than verified service area. This completes a bounded organization-to-funder-to-organization research path, not a peer or prospect ranking.
fetch
ChatGPTRead one ID returned by search. Organization records lack mission/program evidence, so primary or embedded program focus is unverified even when the name suggests one. Funding amounts are historical warehouse aggregates; linked filings are examples for individual funder-recipient pairs.
find_peer_funders
ChatGPTGiven a focal organization EIN and one to five explicitly selected peer EINs, show observed funders across those peers and whether a grant edge is observed for the focal organization. This is a historical comparison, not a recommendation. Hosts that support MCP app views may show a small table; the structured result carries the same rows.
get_research_workflow
ChatGPTRead a workflow ID returned by list_research_workflows. The guide returns reference paths for progressive reading; pass one of those paths as reference to read it. Pass host (claude or chatgpt) to receive a host_adaptation note written for that assistant alongside the neutral instructions. Instructions are not data results or new access. Use the returned connection boundary for saving availability.
list_data_collections
ChatGPTList research collections with live warehouse row counts, table update times, units, limitations, and browsing availability. Table freshness is not source freshness. Use when the user asks about available data or coverage matters to their question; avoid a routine upfront inventory. Returns data for a concise text answer, without an automatic UI panel. Private and operational data are excluded.
list_research_workflows
ChatGPTDiscover the task-focused research guides and the methods they cover. Guides explain how to combine bounded Azimuth reads with native host primary-source research, and list the host profiles (claude, chatgpt) available from get_research_workflow. Internal operations are cataloged but not executable.
map_issue_funding
ChatGPTStart with an exact issue taxonomy ID and state to find candidate organizations and named programs. Discovery returns no funders; use route_basis_filter to inspect organization or named-program routes separately, and candidate_offset for further bounded pages. After checking program evidence, pass selected_eins and purpose_terms to inspect both purpose-mention filing rows and general historical funders of each selected organization. General funding is not attributed to the issue or named program. A purpose mention can be mixed-purpose. Selected EINs without an issue route are labeled. This is not a prospect list.
search
ChatGPTFind organizations by name or EIN and research papers by title/topic. Use a short focused search phrase, then call fetch with a returned ID. Retrieval is for a text answer; no automatic UI panel.
search_interventions
ChatGPTFind intervention taxonomy IDs by topic label or ID before mapping organizations and funding. A related_not_equivalent match is a broader research route, not an exact match for the requested issue; supportive housing currently returns affordable housing this way. A result is not evidence that any organization works on the issue.
