Grant writing desk for small nonprofits: free funder research from public 990-PF grantee lists and Grants.gov, source-tagged proposal drafts that never invent a number, logic models, deadline tracking and donor thank-yous.
Claude Cowork or claude.ai: download the plugin file and install it.
Claude Code:
/plugin marketplace add brianshepardpss/plugin-creator /plugin install grant-desk@plugin-creator
/grant-draft sample
Or, in Cowork or claude.ai, just say: "Try Grant Desk on the sample RFP."
This runs on a fictional organization (Riverbend Youth Literacy, Waco, TX) and a fictional family-foundation RFP (4 questions, word limits, $25,000 cap). The output:
1. AI-policy gate. It reads the RFP's AI-use section first. This one permits AI with disclosure, so it runs in DRAFT mode. 2. Fit memo from real public data. The sample funder is fictional, so the demo looks up a real Waco foundation as a stand-in (Bernard & Audre Rapoport Foundation, EIN 74-2479712). Its 2025 990-PF lists 41 grants totaling $2,020,399. 16 of them, $480,777 in total, went to Waco organizations, and the median comparable grant was $25,000. So the $25,000 cap is a realistic ask there. Every figure comes with the tax year and source URL. 3. Section draft in grants/draft-<funder>.md. Every number is tagged, for example [source: profile#F05]. Gaps such as the demographics of students served are marked [NEEDS DATA: ...], not filled with a made-up statistic. 4. Draft check. A script counts words against each limit, finds numbers with no source tag and numbers that don't match the fact they cite, and warns about stale facts. Sample output is in samples/expected/draft-check.md. 5. Deadlines go to grants/tracker.csv and grants/deadlines.ics (LOI, proposal and both report dates, with reminders 14 and 3 days before each).
The demo uses cached copies of the public data in samples/cache/, so it works offline. On your own funders, the same scripts fetch live data.
| Skill | You say | You get |
|---|---|---|
| grant-draft | "draft this grant", "answer these RFP questions" | AI-policy gate, question map, fit memo, tagged draft, word-limit and number check, deadlines tracked |
| funder-research | "is this foundation a fit?", "who does EIN 75-6015322 fund?", "open federal grants for youth mentoring" | 990/990-PF financials and trend, the grantee list near you, keyword matches, an ask range with its formula, Grants.gov opportunities still open |
| org-profile | "build our org profile", "pull the facts out of our old proposals" | grants/profile.md: mission, programs and a Facts table where every number has a source and an as-of date |
| logic-model-budget | "build the logic model", "write the budget narrative" | a logic model table and a budget narrative that a script checks against your budget CSV |
| grant-tracker | "add the report deadlines from this award letter" | tracker.csv plus a .ics calendar file; relative dates ("6 months after start") computed by script |
| donor-thanks | "year-end thank-you letters from this donor export" | donor segments (monthly, major, first-time, repeat, lapsed) and mail-merged letters, with the IRS $250+ acknowledgment sentence |
| request | "I wish this could..." | a feature request you can file yourself |
Command: /grant-draft [sample | RFP path] [profile path].
ProPublica Nonprofit Explorer API, the GivingTuesday 990 Data Lake (AWS S3) and the Grants.gov API. Nothing about your organization is sent.
working folder and your Claude conversation. Donor data is never sent to any of the services above.
MIT licensed, with its design brief and eval suite: github.com/brianshepardpss/grant-desk. Ideas or problems? Open an issue there, or ask the plugin's request skill to draft one for you.
Not affiliated with or endorsed by ProPublica. Not affiliated with or endorsed by Candid. Not affiliated with or endorsed by GivingTuesday. Not affiliated with or endorsed by Grants.gov or any federal agency. Not affiliated with or endorsed by Amazon Web Services. Not affiliated with or endorsed by the Bernard & Audre Rapoport Foundation, the Meadows Foundation, the Spencer Foundation or NIH.