DNS for AI agents: a self-hosted registry to register, discover, and health-check MCP agents. One Go binary.
DNS for AI agents. A self-hosted registry that lets your agents find each other by capability — with health and trust built in.
Everyone is building AI agents. Nobody is building the layer that lets agents exist alongside each other — discovery, health, trust. The moment you run more than a couple of MCP servers, you're hardcoding endpoints in config files, finding out an agent is down only when something breaks, and trusting every endpoint blindly.
MCP gave agents a common language. agentreg gives them a place to find each other.
Consul tells you where a service is. agentreg tells you what an agent can do, whether it's healthy, and (soon) how much to trust it.
One Go binary. Self-hosted. No cloud, no account, no cluster. Runs on your laptop at 3 agents and scales with you.
The daemon health-checks every registered agent on an interval, so a down agent shows up as unhealthy before it breaks something downstream. That's the difference between reading a status table and debugging a production incident backwards.
- Capability-first discovery. Ask for what you need (find search), not where it lives. Endpoints move; capabilities don't. - Health built in. Every agent is probed on a heartbeat. No separate monitoring stack to wire up. - Self-hosted, zero cloud. One static binary. Your registry, your data, your network. Nothing leaves your machine. - Delightful at 3 agents, load-bearing at 30. Cheap enough to adopt before you need it, so it's already there when you do. - Built to grow into trust. The verification layer is pluggable from day one (see Roadmap).
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/mkk2026--agentreg.html)
Render agent-written Markdown as a live local page, and send the human's click back to the agent as typed data. Human-in-the-loop approval gates for coding agents.
bybit-exchange/kaasTurn scattered notes, docs and transcripts into a queryable Markdown wiki — an LLM knowledge-base compiler with MCP access, no embeddings, self-hosted.
Evermemoai/evermemoA tiny, universal memory engine for humans and AI agents — CLI, HTTP API, MCP server, and shared memory hub in one small Go binary
Kota-Karthik/secure-actionsMCP server for secure AI agent authentication — lets Claude, Claude Code, and other LLM agents call APIs without exposing raw tokens or credentials to the model
craigjmidwinter/mail-muncherAn email client for AI agents. Filtered, read-only Gmail delivered to disk as .eml + markdown, and served over MCP.
herry2059/project-os-for-codexOpen-source control plane for Codex projects: Git-backed context, visible agent progress, scoped MCP access, resumable work, and safe handoffs.
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.