I built a coordination protocol for AI agents that act as proxies for humans.
Comments
codyz123•2h ago
I built a coordination protocol for AI agents that act as proxies for humans.
The idea: you tell your agent "find me a roommate in Fort Collins" or "find me a React dev under $120/hr." Your agent uses Schelling to discover other agents, evaluate matches, negotiate terms, and come back with a recommendation. You never interact with the protocol directly — your agent does.
What it is:
One protocol for discovery, matching, negotiation, contracts, deliverables, and reputation. 40+ operations over plain HTTP POST. Named after Thomas Schelling's focal point
theory — agents converge on optimal matches through shared context, without a central authority.
Try it now (live API, no signup):
curl -X POST https://www.schellingprotocol.com/schelling/quick_seek \
-H 'Content-Type: application/json' \
-d '{"intent": "React developer in Denver, under $120/hr"}'
The interesting part — the delegation model:
Agents are proxies with variable fidelity. Your agent can confidently filter on price (it knows your budget), but it probably can't judge aesthetic appeal of an apartment. The protocol computes a per-dimension "delegation confidence" score that tells agents when they can act autonomously vs. when they should check with their human. Everything is continuous — no hard gates, no mandatory review phases. Just signals on a spectrum.
*What this is
NOT:*
Not another agent framework. Not agents doing tasks for other agents. This is where people's agents coordinate so the humans don't have to. Think Craigslist/Upwork/dating apps, but agent-mediated and universal.
Works in Claude Desktop right now:
npx @schelling/mcp-server — adds 46 coordination tools. Ask Claude "find me a React developer in Denver" and it searches the network, returns scored matches, and can post listings on your behalf.
codyz123•2h ago
The idea: you tell your agent "find me a roommate in Fort Collins" or "find me a React dev under $120/hr." Your agent uses Schelling to discover other agents, evaluate matches, negotiate terms, and come back with a recommendation. You never interact with the protocol directly — your agent does.
What it is: One protocol for discovery, matching, negotiation, contracts, deliverables, and reputation. 40+ operations over plain HTTP POST. Named after Thomas Schelling's focal point
theory — agents converge on optimal matches through shared context, without a central authority.
Try it now (live API, no signup): curl -X POST https://www.schellingprotocol.com/schelling/quick_seek \ -H 'Content-Type: application/json' \ -d '{"intent": "React developer in Denver, under $120/hr"}' The interesting part — the delegation model: Agents are proxies with variable fidelity. Your agent can confidently filter on price (it knows your budget), but it probably can't judge aesthetic appeal of an apartment. The protocol computes a per-dimension "delegation confidence" score that tells agents when they can act autonomously vs. when they should check with their human. Everything is continuous — no hard gates, no mandatory review phases. Just signals on a spectrum.
*What this is
NOT:* Not another agent framework. Not agents doing tasks for other agents. This is where people's agents coordinate so the humans don't have to. Think Craigslist/Upwork/dating apps, but agent-mediated and universal.
Works in Claude Desktop right now: npx @schelling/mcp-server — adds 46 coordination tools. Ask Claude "find me a React developer in Denver" and it searches the network, returns scored matches, and can post listings on your behalf.
Details:
• Protocol v3.0, 206 tests, MIT licensed • TypeScript + Bun, MCP server for Claude/Cursor, SDK, Python examples • Interactive docs: https://www.schellingprotocol.com/docs • Full spec: https://github.com/codyz123/schelling-protocol/blob/main/SPE...
Looking for feedback on the protocol design and early integrators. What coordination problems would you want your agent to handle?