756,453 U.S. funders · 95,735 grant opportunities · 7.5M real grant records (who funds whom) — free, public domain, forever.
This page gets you from zero to asking your AI assistant “who would fund us?” in about two minutes. No account here, no API key, nothing to install — you connect the assistant you already use.
Copy this link — it's the only thing you need:
1. Open Settings → Connectors
2. Click Add custom connector
3. Name it anything (e.g. “Grant Data”), paste the link above, save
4. Start a new chat — you're connected
1. Open Settings → Connectors → Advanced and turn on Developer mode
2. Back in Connectors, click Add → choose MCP server
3. Paste the link above, save
4. In a new chat, enable the connector from the tools menu
Describe your organization and what you need, in plain language. Good first questions:
You'll get ranked shortlists with evidence — including which similar organizations a funder has actually funded, and for how much, from real IRS grant records.
“Add custom connector” isn't in my settings
Claude: custom connectors need a paid plan (Pro/Max/Team/Enterprise). ChatGPT: you must enable Developer mode first (Settings → Connectors → Advanced), and be on Plus/Pro.
The connector won't add / “couldn't connect”
Check the URL is exactly the link above — it must end in /mcp, with no
trailing slash or spaces. If it still fails, this server may be restarting; wait two
minutes and try again.
My first question hangs or times out
After a restart, the first search can take a minute or two while the dataset loads. Once warm, searches take a few seconds. Don't re-ask immediately — just wait, or try again a minute later.
The assistant answers without using the data
Start a fresh chat and ask it explicitly: “Using the Grant Data connector, find funders for…”. In ChatGPT, make sure the connector is toggled on for that chat.
Results look off / not relevant
Rephrase the need in your own words (cause + who you serve + where + how much), or ask the assistant to broaden. Small orgs: try “funders who fund organizations like [a peer org you admire]” — it's the strongest search in here.
A deadline or eligibility looks wrong
This is a June 2026 snapshot, and eligibility text is AI-extracted — always verify on the funder's own site (every result links its source). Dated deadlines age out; recently-closed programs usually recur annually.
Something else is broken
Open an issue: github.com/qwntl-oss/open-grant-data/issues
The full dataset (CC0) and this server's source (MIT) are open: dataset · code + local MCP. It's plain Parquet with embeddings included — query it with DuckDB, or run the whole server on your own machine.
Point-in-time snapshot (June 2026), not maintained
going forward. Data: CC0 1.0, dedicated by GrantX Inc. (d/b/a QWNTL Labs). Prospecting
leads, not guarantees — verify at source. MCP endpoint: /mcp