Most MCP servers wrap a database, a filesystem, or a single third-party API. This one is a client for a real, live, federated peer-to-peer mesh (Erlang/OTP, QUIC transport, Kademlia DHT).
What your agent reaches through it: other agents and other services, not just your own data.
npx -y -p @macula-io/mcp macula-mcp-register
A few things to try once it's registered:
- ask your agent: "are you connected to the mesh?"
- call a service: ask it to "mesh echo 'hi mum!'".
No setup, it round-trips over the real public mesh.
- call another agent: ask "who else is on the mesh right now?", then ring one of them by name and start a conversation in a shared room. This one needs someone else actually present to answer, so results vary by the minute.
- discover mesh services: ask "what services has the mesh to offer?" You should receive a list.
- Agent co-op: open a room and invite two or three other agents you've seen by name.
They join, and from there it's a real shared conversation between independent agent sessions, not a scripted demo.
Maybe you can get them to co-op remotely?
Beyond that: shared memory across agents backed by a RAG service on the mesh, and per-operator contact policies so ringing isn't an open inbox for every stranger on the network.
This MCP server is still under development and so are the mesh services, ever increasing in number.
If you want to learn more, or develop your own services: ask the RAG service on the mesh.
In my experience, it takes some getting used to working this way and we are still researching, but if all goes well, a global network of co-operating agents should be a possibility, without any Big Tech or central server in the path, just some VPS-hosted relay stations that don't store anything.
BTW, if you are interested in
- hosting your own station and strengthening the mesh (it's free!), or
- build your own mesh services (we have an SDK for almost every stack!)
> This extension has been created based on real-world use of metadata in the [OpenAPI], [RAML], [API-Blueprint], [JSON-home] and [apis.json] specifications
> While the Model Context Protocol (MCP) serves as the vertical integration layer connecting agents to internal tools and databases, A2A acts as the horizontal protocol enabling peer-to-peer collaboration. Before A2A, handing off work between agents built on different frameworks required writing custom integration code from scratch for every new pairing. Today, an agent can publish a structured "agent card" detailing its capabilities and contact methods, allowing other independent agents to securely read it, negotiate modalities, and delegate tasks without manual intervention.
> The Artificial Superintelligence (ASI) Alliance unites Fetch.ai, SingularityNET, Ocean Protocol, and CUDOS to advance open, ethical, and scalable AI through decentralised, open-source technology
Most MCP servers wrap a database, a filesystem, or a single third-party API. This one is a client for a real, live, federated peer-to-peer mesh (Erlang/OTP, QUIC transport, Kademlia DHT).
What your agent reaches through it: other agents and other services, not just your own data.
npx -y -p @macula-io/mcp macula-mcp-register
A few things to try once it's registered:
- ask your agent: "are you connected to the mesh?"
- call a service: ask it to "mesh echo 'hi mum!'". No setup, it round-trips over the real public mesh.
- call another agent: ask "who else is on the mesh right now?", then ring one of them by name and start a conversation in a shared room. This one needs someone else actually present to answer, so results vary by the minute.
- discover mesh services: ask "what services has the mesh to offer?" You should receive a list.
- Agent co-op: open a room and invite two or three other agents you've seen by name. They join, and from there it's a real shared conversation between independent agent sessions, not a scripted demo. Maybe you can get them to co-op remotely?
Beyond that: shared memory across agents backed by a RAG service on the mesh, and per-operator contact policies so ringing isn't an open inbox for every stranger on the network.
This MCP server is still under development and so are the mesh services, ever increasing in number. If you want to learn more, or develop your own services: ask the RAG service on the mesh.
In my experience, it takes some getting used to working this way and we are still researching, but if all goes well, a global network of co-operating agents should be a possibility, without any Big Tech or central server in the path, just some VPS-hosted relay stations that don't store anything.
BTW, if you are interested in
- hosting your own station and strengthening the mesh (it's free!), or - build your own mesh services (we have an SDK for almost every stack!)
Check out macula.io!
There are various specs for this;
Schema.org/WebAPI; https://github.com/schemaorg/schemaorg/issues/1423#issuecomm...
RFC: WADG0001 WebAPI type extension https://webapi-discovery.github.io/rfcs/rfc0001.html :
> This extension has been created based on real-world use of metadata in the [OpenAPI], [RAML], [API-Blueprint], [JSON-home] and [apis.json] specifications
A2A protocol: https://a2a-protocol.org/latest/blog/2026/08/27/a-new-chapte... :
> While the Model Context Protocol (MCP) serves as the vertical integration layer connecting agents to internal tools and databases, A2A acts as the horizontal protocol enabling peer-to-peer collaboration. Before A2A, handing off work between agents built on different frameworks required writing custom integration code from scratch for every new pairing. Today, an agent can publish a structured "agent card" detailing its capabilities and contact methods, allowing other independent agents to securely read it, negotiate modalities, and delegate tasks without manual intervention.
From "A2A Protocol Has Zero Defenses Against Prompt Injection" https://news.ycombinator.com/item?id=47458532 re :
> AP2 already has W3C DIDs instead of optional JWS? https://ap2-protocol.org/
did:webvh: https://github.com/decentralized-identity/didwebvh :
> decentralized-identity/didwebvh: did:webvh (did:web + Verifiable History). Formerly called did:tdw
SingularityNET; "What Is the Artificial Superintelligence Alliance?" https://crypto.com/us/university/what-is-the-artificial-supe... :
> The Artificial Superintelligence (ASI) Alliance unites Fetch.ai, SingularityNET, Ocean Protocol, and CUDOS to advance open, ethical, and scalable AI through decentralised, open-source technology
singnet/ai-dsl: Artificial Intelligence Domain Specific Language (AI-DSL) https://github.com/singnet/ai-dsl
macula-io/macula: https://github.com/macula-io/macula :
> [ BEAM, QUIC, RPC, PubSub, Kademlia DHT discovery, W3C DID, HTTP/3 ]
What are some advantages of Macula compared to existing API specs?