Local stdio MCP server for the RigaTech ecosystem — exposes jobs, courses, and events from rinkevich/rigatech-data. Read-only, no authentication.

This is the local counterpart to the hosted server at mcp.riga.tech: same data, same Zod validation, but it runs as a local child process over stdio instead of an HTTP bridge. Use it when you want a self-contained local process, to pin a specific data ref, or to run against a local checkout.

npx -y @rigatech/mcp
# or, with pnpm:
pnpm dlx @rigatech/mcp

It speaks MCP over stdio, so you normally don't run it by hand — your MCP client launches it for you (see below). When run directly it prints ready on stdio to stderr and then waits for JSON-RPC on stdin.

Add this to your MCP client config:

{
  "mcpServers": {
    "rigatech": {
      "command": "npx",
      "args": ["-y", "@rigatech/mcp"]
    }
  }
}

Config file locations:

It can coexist with the hosted server — just give them different keys (e.g. rigatech for this local one and rigatech-remote for npx -y mcp-remote https://mcp.riga.tech).

stack / topicTags / techTags / audienceTags match if the item contains at least one of the given values (case-insensitive). search is a case-insensitive substring match over the title, organization, and English summary.

Each tool returns a structuredContent object { count, items } plus a pretty-printed JSON copy of items in content[0].text.

Pass env vars through your client config:

{
  "mcpServers": {
    "rigatech": {
      "command": "npx",
      "args": ["-y", "@rigatech/mcp"],
      "env": { "GITHUB_TOKEN": "ghp_…" }
    }
  }
}

On the first call to a tool, the server lists the relevant directory in the public rigatech-data repo via the GitHub Contents API, fetches each JSON file from the raw CDN (bounded concurrency), validates every entry with the same Zod schemas used by mcp.riga.tech, and skips anything invalid. Results are cached in memory per dataset for 5 minutes, so repeated calls in a session are instant. The server never writes anything.

git clone https://github.com/rinkevich/rigatech-data-mcp.git
cd rigatech-data-mcp
pnpm install
pnpm build             # tsc -> dist/
pnpm start             # node dist/index.js

pnpm typecheck         # type-check without emitting

# Inspect interactively:
npx @modelcontextprotocol/inspector node dist/index.js

# Run against a local checkout of the data repo (offline):
RIGATECH_DATA_DIR=/path/to/rigatech-data node dist/index.js

Note: stdout is the MCP transport. All diagnostics go to stderr — never write to stdout.

CI (.github/workflows/ci.yml) runs typecheck + lint + build on every push/PR to master. Publishing is automated by .github/workflows/release.yml, which publishes to npm when a vX.Y.Z tag is pushed (or via manual workflow_dispatch).

One-time setup — add an npm automation access token as a repo secret (automation tokens bypass 2FA, which this npm account enforces):

# npmjs.com -> Access Tokens -> Generate New Token -> "Automation"
gh secret set NPM_TOKEN   # paste the token when prompted

Cut a release:

pnpm version patch        # bumps version, creates a commit + vX.Y.Z tag
git push --follow-tags    # pushing the tag triggers the release workflow

The release workflow verifies the tag matches package.json, then runs pnpm publish --access public with npm provenance.

Data lives in github.com/rinkevich/rigatech-data (master). Hosted MCP: mcp.riga.tech. Website: riga.tech.

MIT © Andrejs Rinkevics