SourceHarbor

Proof-first AI knowledge control tower for source intake, grounded search, and agent workflows, with one shared Web/API/MCP truth and strong YouTube/Bilibili lanes.

See It In 30 Seconds · Run Locally · No-Boot Tour · Docs Home · MCP Quickstart · Builders · Distribution Status · Starter Packs · Media Kit · Sample Corpus · Proof · Project Status · Why It Stands Out · Discussions

SourceHarbor helps you turn long-form sources into grounded search results, knowledge cards, traceable job runs, and MCP-ready operations. It stays source-first and proof-first: you can inspect it, run it locally, and verify each surface instead of trusting product copy on vibes alone.

Three quick reasons developers keep reading:

• One truth across Web, API, and MCP. Operators and agent builders see the same jobs, artifacts, and retrieval state instead of three disconnected product shells.
• Proof sits next to the product story. The README, proof ladder, runtime truth, and job receipts all point at the same contract, so the repo earns trust instead of asking for it.
• Strong lanes plus honest boundaries. YouTube and Bilibili are the rich lanes today, RSSHub and generic RSS widen the source universe, and the docs say plainly where route-by-route proof still stops.

The honest intake boundary today is:

• strong support: YouTube channels and Bilibili creators
• general support: RSSHub routes and generic RSS/Atom feeds
• not yet claimable: route-by-route verification for the full RSSHub universe

That intake split now lives behind one shared template catalog instead of separate Web-only presets: the /subscriptions front door, HTTP API, and MCP surface all point at the same strong-supported vs generalized intake contract.

It is strongest when you read it as a control tower for source intake:

• operators use the Web command center
• system builders use the HTTP API
• Codex, Claude Code, and other MCP-aware workflows use the MCP surface
• all three point at the same jobs, artifacts, retrieval index, and operator truth

SourceHarbor is a multi-surface product repo, not a single skill package. Public starter packs and plugin-grade skill surfaces are adoption layers inside that repo. They are not the whole product, and they are not raw exports of the internal .agents/skills tree.

... View full README on GitHub