data-feeder

Declarative data feeds for AI agents. Config file in, MCP server out.

Every agent needs external data. Every developer writes the same fetch/cache/schedule boilerplate for each API. data-feeder replaces that with a YAML config — declare your sources, get a running MCP server with caching, scheduling, and auth.

npx data-feeder init          # creates data-feeder.yaml
# edit data-feeder.yaml
npx data-feeder serve          # starts MCP server on stdio

One feed config:

feeds:
  hn_top_stories:
    source:
      url: https://hacker-news.firebaseio.com/v0/topstories.json
      auth:
        type: none
    cache:
      ttl: 5m
    expose:
      type: tool
      name: get_hn_top_stories
      description: "Get the top story IDs from Hacker News"

What the agent sees: a tool called get_hn_top_stories that returns cached HN data, automatically refreshed every 5 minutes.

• YAML config — declare data sources, no code required
• Caching — per-source TTL, in-memory, with hit rate stats
• Scheduled prefetch — cron expressions or simple intervals
• Auth — API key (header or query param), Bearer token, x402 payment protocol
• MCP tools & resources — auto-generated from config
• Hot reload — edit config while running, feeds update without restart
• Budget enforcement — daily limits and per-call caps for paid APIs
• Retries — exponential/linear backoff, respects Retry-After
• Response extraction — dot-notation path to pull nested data

npx data-feeder init                     # scaffold config + .env.example
npx data-feeder serve                    # start MCP server (stdio)
npx data-feeder serve --http 3100        # start on HTTP port
npx data-feeder validate                 # check config for errors
npx data-feeder validate --config ./alt.yaml
npx data-feeder list                     # list all configured feeds