A Claude Code skill that finds non-dilutive money for a startup and helps win it: grants, cloud and AI credits, vouchers, public programmes, accelerator perks. Any country. It interviews the founder first, searches on its own, scores what it finds, and then does the paperwork with the founder's approval on every submission.
The playbooks were written from applications the author actually filed, not from listicles: a Stellar Community Fund build award, a French Tech Ticket laureate slot funded through Bpifrance, an Estonian innovation-agency dossier in progress, and the rejections in between. The lessons that repeat across funders are the point of the repository.
git clone https://github.com/tuturama/grantsmith ~/.claude/skills/grantsmithThen, in any Claude Code session, /grantsmith or simply "find grants for my company". Claude loads the skill when the task matches.
Four phases, in order. The first one is never skipped.
1. Intake. A conversation, not a form. Legal entity and country of incorporation (eligibility is tied to the entity, not to where the founder sits), company age, ownership, stage, sector in the funder's vocabulary, what the money is for, and how many founder-hours are available. The result is written to funding-profile.md in your project and reused from then on.
2. Search. Autonomous, in layers, from fastest to slowest money: cloud and AI credits, the national innovation agency, regional and municipal programmes, sector and foundation grants, accelerators and competitions. Aggregators are used to find names; eligibility, amounts and deadlines come from the funder's own page. When a page blocks fetching, a scraping ladder climbs one rung at a time.
3. Evaluate. Hard gates first (entity country, company age, size class, sector codes, registry hygiene). What survives is scored on fit, amount and probability, divided by effort, and returned as a ranked shortlist of about seven, deadlines inside thirty days on top. For each option: what it is, why it fits this company, what it takes, when it closes.
4. Execute. A reusable answer kit is built once (one-liner at three lengths, problem and solution, traction with numbers, team, budget blocks) and tailored per application, never pasted blind. Forms are filled with browser automation, with the known traps written down. Outreach goes only to addresses verified on an official page. Every application gets a pipeline entry, a dated log line and a follow-up date.
- Nothing is submitted or sent without the founder seeing the final text and saying yes.
- Nothing is fabricated. A field the profile cannot answer is a question for the founder.
- Names, addresses and numbers must match across the application, the company registry and the billing accounts. Mismatches are a silent rejection reason at credit programmes.
- CAPTCHAs and identity checks are the founder's to click. The skill never bypasses bot detection.
- Deadlines are data. They are recorded the moment they appear and surfaced at the top of every status update.
> find grants for my company
Before I search, a few questions. Where is the company incorporated, and when?
…
Profile saved to funding-profile.md. Searching in five layers.
Shortlist (7 of 23 found, hard gates applied):
1. <national agency> Development Grant, up to €35k at 70%, rolling. Fits: R&D on <your tech>. Effort: 2 to 3 founder-days plus a credited provider. Next: pre-consultation email to the agency.
2. <cloud provider> startup credits, ~$5k self-serve. Fits: current infra bill. Effort: 30 minutes. Next: apply, billing account must be on the company domain.
…
Closing within 30 days: none. Pick the ones to pursue and I will build the answer kit.
SKILL.md the skill: four phases and the hard rules
references/
intake.md the founder interview
search-playbook.md the five layers, query patterns, source hygiene
sources.md programme families and the lessons that repeat
credits-catalog.md cloud, AI and SaaS credit programmes (starting points)
country-programs.md agencies and flagship programmes by region (starting points)
evaluation.md hard gates, scoring, shortlist discipline
scraping-tools.md fetch, then Firecrawl or Jina, then a browser, then a clean IP
forms-playbook.md form automation and WAF or Turnstile troubleshooting, in order
application-strategy.md answer kit, tailoring, outreach, follow-ups
The two catalogues are starting points, not directories. Programmes rename, merge and expire every quarter; the durable knowledge is the agency name. Every entry is verified on the funder's current page before it reaches the founder.
- It does not guarantee money. It finds what you are eligible for and makes the application as good as your facts allow.
- It does not write your traction. If there is nothing live yet, the answer kit says so, and the shortlist shifts toward credits and pre-revenue programmes.
- It does not replace a grant consultant for a two-million-euro EU consortium bid. It gets you to the pre-consultation with the right agency and a clean dossier.
Additions welcome, in three kinds: a programme family or agency that is missing from a catalogue (with the official URL), a lesson that repeated for you across funders, or a trap in a form that cost you an hour. Keep entries verifiable and dated. See the note at the top of each catalogue before adding to it.
MIT. Made by Tuturama, an AI-native software factory for systems that cannot break.