find_nonprofits
ChatGPTResolve nonprofit organizations by name or EIN — fast, no analysis. Pass a list of names and/or EINs (mixed is fine). For each NAME you get a short RANKED CANDIDATE list, best match first — many organizations share similar names, so candidates are deliberate, not errors: pick the best one yourself, preferring the expected location. EINs are validated and looked up exactly. To sharpen name matching, pass state (two-letter code) and/or put location words in the name itself ("Second Harvest Spokane"). Each candidate includes name, EIN, location, directory revenue, ratings, a one-line mission summary, and the organization's DATA RECENCY — its latest 990 filing (IRS tax year, fiscal period end, form type, date received) and latest parsed impact-report year, or their stated absence — so you can judge how current its data is. Resolve names to EINs here before research_nonprofit_data when precision matters.
get_grants_made
ChatGPTGrants one organization reported making on one filing — exact rows, no analysis. Pass its 9-digit EIN; optionally tax_year (default: the newest year that has grants), page (from 1) and limit (default 50, max 250). The reply names the year and filing served, the other years with grants, the row count and page math, and the year's totals split into paid and approved for future payment. Rows, largest first: grantee name as filed and its directory name when the EIN is on file (an unnamed grant with is_to_organization false went to an individual), grantee EIN, location, amount, cash and non-cash amounts, purpose, status, and the 990 schedule it came from. Every status is included. For anything beyond these fixed facts — comparisons, screens, trends, calculations, or any other field — use research_nonprofit_data.
get_grants_received
ChatGPTGrants one organization received, as reported in its funders' filings for one funder tax year — exact rows, no analysis. Pass its 9-digit EIN; optionally tax_year (the funders' filing year; default: the newest with rows), page (from 1) and limit (default 50, max 250). The reply names the year, the other funder years with rows, the row count and page math, and the total for the set. Rows, largest first: funder EIN and directory name, location, amount, purpose, status, and the funder's form type and filing date. Only grants whose funder filing carries this EIN are here; grants reported by name alone are not, so the total is a floor. For anything beyond these fixed facts — comparisons, screens, trends, calculations, or any other field — use research_nonprofit_data.
get_key_financials
ChatGPTEvery 990 filing on record for one organization, newest first — exact headline figures, no analysis. Pass its 9-digit EIN. One entry per filing: tax year, form type (990, 990-EZ, 990-PF), the period's begin and end dates (a period well under a year is a fiscal-year- change stub), filing date, total revenue, total expenses, total assets, total liabilities, net assets, contributions and grants, program service revenue, investment income, other revenue (those four are null on a 990-PF), employees and volunteers (full 990 only). Not paged: the whole history returns. Figures are as filed. For anything beyond these fixed facts — comparisons, screens, trends, calculations, or any other field — use research_nonprofit_data.
get_nonprofit
ChatGPTOne organization's profile — exact facts, no analysis. Pass its 9-digit EIN (resolve a name with find_nonprofits). Returns, from the Giving Compass directory: name and alternate name, address, website and donation link, founding year, IRS ruling date, principal officer; classification (501(c) subsection, foundation type, NTEE code, issue area, category, group affiliation, deductibility, active or inactive, GC vetted); Charity Navigator score and stars; the summary and impact summary; the directory's latest revenue, expenses, net assets and employee count with their filing year and source; the tax years on file for each drill-down (filings, grants made, grants received, people, programs); and its parsed impact-report evidence in full — the report years, every metric and every story, newest first. Impact-report coverage is limited and skews to vetted and larger organizations, so its absence is normal. An EIN outside the directory returns found=false with whatever is on file for it. For anything beyond these fixed facts — comparisons, screens, trends, calculations, or any other field — use research_nonprofit_data.
get_people
ChatGPTThe officers, directors, trustees, key employees and highest-paid staff one organization listed on one filing (Form 990 Part VII) — exact rows, no analysis. Pass its 9-digit EIN; optionally tax_year (default: the newest year with people listed), page (from 1) and limit (default 50, max 250). The reply names the year and filing served, the other years available, the row count and page math. Rows, highest total compensation first: name, title, role flags, compensation from the organization, from related organizations, other compensation, and hours per week. Unpaid board members appear with zero compensation. For anything beyond these fixed facts — comparisons, screens, trends, calculations, or any other field — use research_nonprofit_data.
get_programs
ChatGPTThe program services one organization described on one filing (Form 990 Part III) — exact rows, no analysis. Pass its 9-digit EIN; optionally tax_year (default: the newest year with programs; a 990-PF lists none), page (from 1) and limit (default 50, max 250). The reply names the year and filing served, the other years available, the row count and page math. Rows in filing order: the description as written, expenses, grants included in those expenses, and revenue. For anything beyond these fixed facts — comparisons, screens, trends, calculations, or any other field — use research_nonprofit_data.
research_nonprofit_data
ChatGPTRun a substantive nonprofit-data analysis over Giving Compass's warehouse: every US IRS-registered nonprofit, with EVERY field of every Form 990, 990-EZ, and 990-PF as filed (full balance sheets, all schedules, officers, grants, programs), multi-year, plus classifications, ratings, parsed impact reports, and census joins. Use this for anything needing computation or synthesis: counts and screens, rankings, comparisons, financial trends over years, sector aggregates — and deep analytical work: balance-sheet reconciliations, related-organization tracing, testing a screen's precision across its cohort. Send ONE analytical question per call, stating exact criteria where you have them (geography, thresholds, years, EINs — resolve names with find_nonprofits first). Continuity: every reply carries this analysis's analysis_id. To refine or continue the analysis, pass the analysis_id back and state ONLY what to change — every other declared choice is preserved exactly. You can also ask QUESTIONS of an existing analysis — verify a figure, explain a row — answered without changing its declared state. Omit analysis_id to start a new, unrelated analysis. Ids come only from prior replies (an unknown id errors); never invent one. Where your request leaves room for interpretation, the analyst makes judgment calls and DECLARES every one in the reply's analysis manifest — if an assumption doesn't match your intent, send a refinement; your specification wins. Takes materially longer than the lookup tools (typically 15-90 seconds). Returns a markdown report: exact figures, tables and chart data, organization lines with EINs, the full analysis manifest (the declared state), what changed on a refinement, and a closing audit with the exact queries run — use it to verify (ask a follow-up when a figure doesn't reconcile).
search_articles
ChatGPTSearch Giving Compass's curated editorial articles on philanthropy and social impact by topic. Use for sector commentary, donor guidance, and issue-area context — supplementary reading, not primary data (organizational data comes from the other tools). time_scope "recent" (default) covers the past year plus evergreen pieces; "all" searches everything. Returns titles, publishers, dates, excerpts, and links. The archive spans years and some pieces are older — every result shows its date; weigh recency when the topic is time-sensitive.
find_nonprofits
Claudeget_grants_made
Claudeget_grants_received
Claudeget_key_financials
Claudeget_nonprofit
Claudeget_people
Claudeget_programs
Clauderesearch_nonprofit_data
Claudesearch_articles
Claude