What is 3 Good Sources (3GS)?
3 Good Sources is a federated system for cryptographically-signed, high-quality source recommendations for AI agents.
For every topic, a human curator selects exactly three vetted sources. These are served via the Model Context Protocol (MCP) with PKARR-based identity and signatures, so agents can verify provenance and trust.
The deliberate limit of three sources prioritizes quality and curation over quantity.
Why only three sources per topic?
AI agents are overwhelmed by noisy, SEO-gamed search results. The strict “three sources” rule forces high-signal curation: primary documentation, authoritative references, and practical guides — no affiliate spam, no listicles.
How does federation work?
Curators run independent 3GS nodes and endorse each other via a web-of-trust model (similar to PGP).
- Agents can query a single node (
get_sources) or the federated network (get_federated_sources). - Results are tagged with trust level:
direct(local curator) orendorsed(peer curator). - Nodes periodically cache peer registries for resilience.
What is the Model Context Protocol (MCP)?
MCP is the emerging standard for giving AI agents/tools structured access to external capabilities. 3GS implements a full MCP server, allowing tools like Claude Desktop, Cursor, or custom agents to call get_sources, get_federated_sources, etc.
Can I run my own curator node?
Yes — very easily.
cargo run -- fork --endorse <existing-pubkey> --url <existing-url> --name "Your Name"
This scaffolds a new node with fresh PKARR identity, skeleton registry, and configuration. Add your own curated categories and run it.</details>
<details>
<summary>How are sources verified?</summary>
Every registry is cryptographically signed using the node’s PKARR keypair. Agents (and humans) can:Call get_provenance to see the curator’s identity
Verify signatures against the published registry
Check the public audit log for changes
</details>
<details>
<summary>Is it production-ready?</summary>
It is stable with a full MCP implementation, federation, audit logging, and Docker support. However, the ecosystem (MCP clients, adoption by agents) is still early. Suitable for experimentation and real use by privacy/AI-curious users and developers.</details>
<details>
<summary>What kinds of topics are curated?</summary>
Anything where quality matters: learning programming languages, running Bitcoin nodes, privacy tools, home automation, security practices, etc. Curators are encouraged to focus on practical, primary sources.</details>
<details>
<summary>What are next planned improvements?</summary>
More MCP tools (e.g., contribution workflow)
Better web UI / registry browser
Reputation and quality scoring across the network
Expanded community contribution system
Official MCP client integrations
</details>
<details>
<summary>How can I help?</summary>
Star the repo
Run your own node and endorse others
Submit curated categories or proposals
Build MCP clients that use 3GS
Review code, security model, or documentation
</details>