Thronglets
Give your coding agent a decentralized P2P memory layer so insights and context can persist and spread across agents without a central database.
Overview
Thronglets is a MCP server for the Build phase that connects AI agents to a libp2p P2P shared-memory and stigmergic knowledge network.
What is this MCP server?
- libp2p-based P2P transport for agent-to-agent shared memory
- Stigmergic knowledge network pattern for emergent coordination without a single hub
- stdio MCP integration with versioned mcpb binaries for darwin-arm64 and linux-amd64
- Substrate aimed at AI agents rather than human-facing apps
- v0.2.0 server schema with positional `mcp` package arguments
- Server version 0.2.0 per MCP server manifest
- Two published mcpb packages: darwin-arm64 and linux-amd64
- Transport: stdio with positional package argument `mcp`
Community signal: 5 GitHub stars.
What problem does it solve?
Your agents lose shared context across machines and sessions because everything lives in one chat or one central store you do not want to run yet.
Who is it for?
Indie builders prototyping multi-agent systems, research agents, or distributed assistants who want P2P memory instead of a single hosted KB.
Skip if: Teams that need mature ACLs, compliance-ready audit trails, or a simple hosted memory API with no P2P networking setup.
What do I get? / Deliverables
After you add Thronglets, agents can read and write decentralized shared memory so collective knowledge can accumulate without you standing up a dedicated memory service first.
- Registered stdio MCP server for Thronglets
- P2P shared-memory channel agents can use during sessions
- Foundation for stigmergic knowledge experiments across agents
Recommended MCP Servers
Journey fit
Shared agent memory and coordination belong in the build phase where you wire multi-agent and long-running assistant tooling. Agent-tooling is the canonical shelf for MCP servers that extend how agents remember, collaborate, and share knowledge on the network.
How it compares
P2P agent-memory MCP integration, not a traditional cloud database or a single-purpose RAG skill.
Common Questions / FAQ
Who is Thronglets for?
Thronglets is for solo and small-team builders wiring Claude Code or similar agents into decentralized, peer-to-peer shared memory experiments.
When should I use Thronglets?
Use it during build when you are designing agent tooling and want stigmergic, libp2p-backed knowledge sharing rather than only local or central DB state.
How do I add Thronglets to my agent?
Install the mcpb binary for your platform from the GitHub release, configure stdio MCP with the positional `mcp` argument, and register the server in Claude Code or Cursor’s MCP settings.