US Nonprofit Grants Awarded
U.S. · 6,465,185 grant awards
Every grant reported on a US nonprofit tax return: 6,465,185 grant records totalling over $680 billion, parsed from Form 990-PF and Form 990 Schedule I. Each record names the funding organisation by EIN, the recipient, the cash amount and the stated purpose.
More about this list
This is the file that answers “which foundations fund organisations like mine”, and it is not sold as a bulk download anywhere else.
Need just some of them? Ask in your own words.
≡Sample Data Preview
| funder ein | recipient | recipient ein | amount | purpose |
|---|---|---|---|---|
| 616168361 | NORTON HEALTHCARE SYSTEM | 376672 | GENERAL SUPPORT | |
| 852275972 | Frank Lloyd Wright Foundation | 7500 | Scholarship | |
| 844595898 | Harkham Hillel Hebrew Academy | 1000 | TUITION SUPPORT | |
| 208057969 | HEDLUND KEITH | 341 | HOME FURNISHINGS | |
| 450931286 | RONALD MCDONALD HOUSE CHARITIES OF NEW MEXICO | 850283204 | 11129 | GENERAL SUPPORT |
{}What's in each row
| Field | What it is | Rows filled |
|---|---|---|
| funder_ | EIN of the organisation that awarded the grant | 100% |
| recipient | Recipient organisation name as written on the return | 100% |
| recipient_ | Recipient EIN where the form required one (Schedule I) | 35% |
| amount | Cash grant amount | 100% |
| purpose | Stated purpose of the grant | 99% |
Rows filled is measured on every row of the delivered file.
✓Where this data comes from
This dataset is built from the IRS's own public files — the Business Master File of exempt organisations, the Form 990/990-EZ/990-PF e-file XML, the 990-N e-postcard file, the auto-revocation list and Publication 78 — joined on EIN, with the Federal Audit Clearinghouse for single-audit detail and the US Census geocoder for coordinates. Nothing is collected from an organisation's own website: every field traces to a filing the organisation made or a determination the IRS published.
September 9, 2026
File built
5 fields
Columns in the file
Fill measured
On every row of the delivered file
Records are published as the IRS holds them — nothing is silently dropped for looking wrong, because on this file an empty field is itself information (a church that files no return is not a bad record). Coordinates are validated to US bounds, ZIPs and phone numbers to format, and every organisation carries an activity_status measured from both filing channels — the 990 e-file index and the 990-N postcard file — rather than inferred from registration. Fill rates vary sharply by slice and are stated on each dataset.
How we build and date every file: methodology.
Using ChatGPT, Claude or an API?
No setup needed. Ask ChatGPT or Claude your question about the US Nonprofit Grants Awarded and give it this page: it can open https://locationlists.com/find?dataset=us-nonprofit-grants-awarded with the filters you need and show you the count and price.
Using ChatGPT or Claude? Tell it to answer from locationlists.com/find — no account, no connector.
How assistants ask
No connector needed. Any AI assistant can answer from this site with a web link, no connector, no account: open https://locationlists.com/find with the question as query parameters.
If a question needs filtering, distance, drive time, combining lists or a price, open a /find link: either put the question itself in q=, or build the structured link. The answer does not need an API, a connector or an account.
Simplest, the question itself: https://locationlists.com/find?q=<the question in plain words>
Or set the parts yourself: https://locationlists.com/find?dataset=<slug>&state=<ST>&city=<city>&county=<county>&zip=<zip>&near=<place or zip>&radius=<miles>&drive=<minutes>&where=<column>:<op>:<value>&total=<rows>
Leave out what the question does not need. Use radius for straight-line miles or drive for minutes of driving (5 to 60), not both; near can name several places separated by | to reach from any of them. To ask about several lists at once, use datasets with two or more names, or category with a kind of business. To compare two lists, add relate=nearest and the other list with b. in front of its parameters. To count by area, use areas=county (or zip, state, metro). To keep only places whose county, ZIP code, metro area or state meets a Census figure, add area_where (for example county:population>1000000); with areas, per=100000 gives counts per 100,000 residents.
Links an assistant can open, each showing how many rows match, a few preview rows and the price for just that part:
- Only rows where amount is over 7500: https://locationlists.com/find?dataset=us-nonprofit-grants-awarded&where=amount%3Agt%3A7500
- The whole list: add a city, state, ZIP code or distance:
https://locationlists.com/find?dataset=us-nonprofit-grants-awarded
Optional: connect for repeat use or wallet payments:
Read https://locationlists.com/skill.md and follow the instructions to connect LocationLists. Then show me the US Nonprofit Grants Awarded list.Your assistant will walk you through setup and show you the price before you pay.
For developers
Every step below has a no-connector equivalent: https://locationlists.com/find?dataset=us-nonprofit-grants-awarded with the same filters as query parameters.
- Connect your agent to
https://locationlists.com/mcp(no key, no signup). - Call
count_locationswithdataset: "us-nonprofit-grants-awarded"and any filter — it is free, and returns how many rows match and what they cost. - Fetch the rows with
query_locations: $0.05 per 100 rows from this dataset. Above 3,232,593 rows, buying the whole file is cheaper.
Paid in USDC over x402 — the agent gets an HTTP 402 with the exact price, signs, and retries. Money moves only after rows are delivered, so a failed or empty query costs nothing. Setup for Claude, Cursor and VS Code
?Frequently Asked Questions
Is this the same data as GuideStar or Candid?
It comes from the same public IRS sources Candid uses — they list them on candid.org/about/our-data/data-sources. Against Candid's published Essentials API schema we match 41 of the 66 fields it returns, which is 84% of the fields reproducible from public records. What we do not have is the part organisations submit to Candid directly: board rosters, DEI demographics, Seals of Transparency, awards and photos. If you need those, Candid is the right purchase. If you need the whole file as a CSV, this is.
Why is this so much cheaper than a Candid API subscription?
Candid sells a subscription starting at $4,800 a year. This is a one-off purchase — and you are not giving up API access to get the lower price. The same 6,465,185 records are available three ways: the bulk CSV, our MCP server for programmatic queries, and per-record pricing so an application or AI agent can pull just the rows it needs. What you are not paying for is an annual seat.
Do you have an API, or only CSV downloads?
Both. Every dataset is exposed through our MCP server — search, sample, quote, buy and query — and priced per record, so you can pull a single EIN or a filtered slice without buying the file. Payment is by card or stablecoin with no account to create. The CSV is simply the cheapest route once you want more than about half the rows.
How do I know an organisation on this list still exists?
Every row carries activity_status. “filing” means the organisation filed a 990, 990-EZ, 990-PF or 990-N postcard within the last two tax years. “revoked” means it appears on the IRS auto-revocation list with no later determination letter. Registration alone is not treated as evidence of life — 435,906 organisations in the full IRS file have no filing on record at all.
What is missing, and can you add it?
Deep Form 990 financial detail (every line of Parts VIII–X), websites and logos for organisations that did not report one, and subject/population classification beyond the IRS NTEE code. All three are buildable — the 990 line items are in files we already hold. Choose “Custom enrichment” at checkout and we will scope it with you.
Can an AI agent buy a slice of this instead of the whole file?
Yes. The dataset is exposed through our MCP server with per-record pricing, so an agent can query and pay for exactly the rows it needs rather than the whole download.