MCP server and CI for grants

GrantKit is designed to be driven by agents and by continuous integration. Both surfaces use the same stateless engine and return the same JSON.

MCP server¶

Expose the engine to an AI agent over the Model Context Protocol. Install the mcp extra and run the console script (stdio transport):

pip install "grantkit[mcp]"
grantkit-mcp

Tools¶

Tool	Arguments	Returns
`grant_check`	`path`	The `checks` structure (`errors`, `warnings`, `items`).
`grant_status`	`path`	The full `status.json` document.
`grant_build`	`path`, `format`	Output paths plus the `status.json` document.

Each path is a directory containing grant.yaml. The server makes no AI or network calls.

Registering with Claude Code¶

{
  "mcpServers": {
    "grantkit": { "command": "grantkit-mcp" }
  }
}

CI for grants¶

A composite GitHub Action lints every push and publishes the review page and status.json as build artifacts:

# .github/workflows/grant.yml
name: Grant
on: [push, pull_request]
jobs:
  grantkit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: GrantKit/grantkit@v0.2.1
        with:
          path: .          # directory containing grant.yaml
          strict: "false"  # set "true" to fail on warnings too

The action runs check --json (captured, non-fatal), build --share, and status --json, uploads assembled.html, status.json, and check.json, and then enforces the result — errors always fail the job; warnings fail it only under strict: "true".

The status.json contract¶

Both surfaces produce the same artifact the CLI does. Its exact shape — the stable contract other tools depend on — is documented in artifacts.