Find AI Grants with Claude and ChatGPT: How AI Assistants Use Our Database

Most grant seekers still search for funding the same way: open a browser, type keywords into a portal, scroll through hundreds of results, click into each one, copy details into a spreadsheet. Repeat for the next agency. And the next.

What if your AI assistant could do that entire workflow for you -- querying a real grant database, pulling funder profiles, checking deadlines, and analyzing competitive landscapes -- all inside the same conversation where you are drafting your proposal?

That is now possible. Granted operates a Model Context Protocol (MCP) server that connects directly to Claude Desktop, Cursor, and any other MCP-compatible client. No API key. No account. No cost. You add five lines of JSON to a config file, restart your app, and your AI assistant gains access to 87,000+ grants, 133,000 foundation profiles, and federal award history.

What MCP Actually Does

The Model Context Protocol is an open standard created by Anthropic that lets AI assistants connect to external data sources in real time. Think of it as a USB port for AI: instead of the model relying solely on its training data or web search, MCP lets it call specific tools hosted by third-party services.

When you connect Claude Desktop to Granted's MCP server, Claude does not just search the web and summarize what it finds. It queries Granted's actual database -- the same one that powers grantedai.com -- using structured tools that return scored, ranked, and verified results. The difference is the same as asking a friend to Google something versus giving them direct access to the source system.

Granted's MCP endpoint supports both Streamable HTTP and legacy SSE transports, so it works with Claude Desktop, Cursor, Windsurf, and any client that implements the MCP specification.

Setting Up Claude Desktop in 30 Seconds

Open your Claude Desktop config file (claude_desktop_config.json) and add the following:

{
  "mcpServers": {
    "granted": {
      "type": "url",
      "url": "https://grantedai.com/api/mcp/mcp"
    }
  }
}

Restart Claude Desktop. That is the entire setup. You should see Granted listed as an available MCP server, and Claude can now call its tools as part of any conversation.

There is no authentication step, no account to create, and no usage fees. Every tool is read-only -- the server cannot modify data or take actions on your behalf.

The Seven Tools Your AI Assistant Gets

Once connected, Claude (or any MCP client) can call seven tools depending on what you ask:

search_grants is the primary discovery tool. It runs your query against Granted's database using full-text search, embedding similarity, and a 15-feature scoring algorithm -- the same pipeline that powers the grant search on grantedai.com. Results come back scored and ranked, with titles, deadlines, amounts, eligibility, and direct links.

search_ai_grants is a specialized variant pre-filtered for artificial intelligence and machine learning funding. If you are looking for AI-specific opportunities -- NSF programs, DARPA BAAs, DOE computing solicitations -- this tool skips the noise and returns only AI-relevant results.

get_grant pulls the full detail page for any grant by its slug: description, eligibility requirements, award amounts, application links, deadlines, and source metadata.

search_upcoming_deadlines finds grants closing within a timeframe you specify. Ask for grants due in the next 30 days, 60 days, or any window. Useful for building a submission calendar or catching opportunities you might have missed.

search_funders searches across 133,000+ foundation profiles drawn from IRS 990 filings. Filter by name, state, NTEE code, asset size, or annual giving. If you are exploring private foundation funding, this is the fastest way to build a prospect list.

get_funder returns a full foundation profile: financials, recent grants, officers, geographic focus, and giving patterns.

get_past_winners pulls federal grant recipient data. See who won awards from a specific agency, how much they received, and where they are located. This is competitive intelligence for positioning your own proposal -- knowing who funds what, and at what scale, tells you whether your organization fits.

What This Looks Like in Practice

The real power is that these tools compose naturally inside a conversation. You do not need to call them individually or remember their names. Just ask questions in plain language.

A few examples:

• "Find NIH grants for neuroscience research with deadlines in the next 60 days." Claude calls search_grants with your query and search_upcoming_deadlines filtered to 60 days, then merges the results.

• "Which foundations in California give more than $1M annually to environmental causes?" Claude calls search_funders with state, giving threshold, and NTEE filters.

• "Show me who won NSF CAREER awards in computer science last year and how much they received." Claude calls get_past_winners with NSF and CAREER parameters.

• "I'm a small nonprofit in rural Appalachia. What USDA grants am I eligible for?" Claude calls search_grants with your eligibility context baked into the query, then filters results by organization type.

You can chain these into multi-step research sessions. Start with a broad search, drill into a specific grant, look up the funder's profile, check who else has won similar awards, and then ask Claude to outline how your organization might position a proposal -- all without leaving the conversation.

Why This Matters for Grant Seekers

The grant discovery problem has never been about the existence of opportunities. Federal agencies, state programs, and private foundations collectively distribute hundreds of billions of dollars annually. The problem is finding the right opportunities for your specific organization, at the right time, with enough lead time to prepare a competitive application.

MCP changes the dynamic because it removes the context-switching that slows down research. Instead of bouncing between Grants.gov, foundation databases, and award registries -- copying and pasting into your notes as you go -- you stay in one place. The AI assistant handles the retrieval while you focus on strategy.

Granted's MCP server also shares its cache with the main website. Queries run through MCP warm the same cache that serves grantedai.com, and vice versa. That means frequently searched topics return results in seconds, and the database gets richer over time as more people search.

For teams that already use Claude Desktop or Cursor for writing, the integration is immediate. The same assistant helping you draft a specific aims page can now pull real grant data, check whether a deadline is still open, and look up whether your funder has a history of supporting similar work -- all within the same thread.

Full documentation, the complete tool reference, and example queries are on the Granted MCP page. For researchers and nonprofits working in artificial intelligence specifically, the AI grants category tracks current federal and private opportunities across NSF, NIH, DARPA, DOE, and more.

The gap between finding a grant and writing a competitive proposal is where most applicants lose time, and connecting an AI assistant directly to a live funding database is one way Granted is closing it.