sourceharbor

Why SourceHarbor Stands Out

SourceHarbor is not trying to be “another summarizer repo.”

It is stronger when you read it as a productized knowledge pipeline.

Comparison Matrix

Capability	Transcript-only tool	Summary-only script	Internal dashboard	SourceHarbor
Continuous source intake	No	No	Partial	Yes
Job-level trace and retries	No	No	Partial	Yes
Digest feed for operators	No	Partial	Yes	Yes
Retrieval over generated artifacts	No	Partial	Partial	Yes
Notifications and digest delivery	No	Partial	Partial	Gated
MCP surface for agents	No	No	No	Yes
Public proof path	Rare	Rare	Rare	Yes

Notifications and digest delivery are implemented, but live send claims still depend on sender identity, mailbox, and provider readiness.

The Differentiator

Most repositories stop after “generate a summary.”

SourceHarbor continues:

1. Capture sources repeatedly through a source-universe intake front door
2. Run a job-backed pipeline
3. Write artifacts you can inspect later
4. Search those artifacts
5. Track repeated themes as watchlists instead of ad-hoc tabs
6. Merge repeated runs into trends, briefings, and a shared Ask story payload
7. Deliver them through notifications
8. Reuse the same system through MCP

Front-Door Line

The product line is stronger when you read the doors in order:

1. Subscriptions widens source intake honestly
2. Watchlists saves the tracking object
3. Trends becomes the compounder front door
4. Briefings lowers cognitive load with one current story
5. Ask carries that story context into answer, changes, and evidence
6. MCP reuses the same truth for Codex, Claude Code, and other agent clients

Trade-Offs

SourceHarbor is a better fit when you want:

• a durable intake-to-artifact flow
• operator visibility
• agent integrations
• a source-first system you can inspect deeply

It is a worse fit when you want:

• a tiny copy-paste summarizer snippet
• a zero-infrastructure hosted tool
• a repo with no operator or orchestration layer

Why This Matters For GitHub Discovery

This differentiation gives the repo a stronger public label:

SourceHarbor is the GitHub-native control tower for long-form knowledge pipelines.

That label is easier to remember, easier to share, and easier to star than a vague “AI information hub” description.

Ecosystem Fit

Ecosystem	Fit level	Why
MCP	Primary	real, shipped surface today through ./bin/dev-mcp and apps/mcp/server.py
Packaged public CLI	Primary	packages/sourceharbor-cli is now a thin installable wrapper that delegates into repo-local bin/sourceharbor when a checkout is present
Public TypeScript SDK	Primary	packages/sourceharbor-sdk now exposes the same typed HTTP client/url/type substrate that the web app uses
Public starter surface	First-cut	starter-packs/ is the public entry directory, with docs/public-skills.md, docs/compat/*, templates/public-skills/*, and examples/* acting as companion starter assets
Codex	Primary	strong fit for source-first local workflows that want to use MCP or HTTP against the same operator truth
Claude Code	Primary	same fit pattern as Codex: local, MCP-aware, API-capable, and proof-first
OpenHands	Secondary / comparison	adjacent as an agent-runtime ecosystem, but SourceHarbor is not a generic software-task agent
OpenCode	Secondary / comparison	adjacent as a coding/automation workflow surface, but not a primary product identity here
OpenClaw	First-cut local pack	the shared MCP / HTTP substrate is real and the repo now ships a local starter pack, but it still should not be sold as a SourceHarbor plugin-first identity or primary front door

Here, First-cut means the starter layer is real and usable today, but it is still a public entry surface rather than a fully hardened standalone ecosystem product.

This site is open source. Improve this page.