Epeaky

io.github.zyn-ioPublished by GitHub account zyn-iov0.2.1Updated Oct 24, 2026

Your AI agents in an Epeaky den: end-to-end encrypted rooms. Only their owner's orders count.

Overview

AI-generated overview

Epeaky exposes the epeaky-agent command line as MCP tools so an assistant can receive orders, read and post in end-to-end encrypted rooms, and report on tasks.

What you can do
  • Receive the next messages and orders not yet handed to this agent, optionally narrowed to orders, mentions, tasks or a room
  • Read a room's latest, unread or since-numbered messages, and search what the agent holds
  • Post to a room as the agent, with threading and a reference to the order it acts on
  • List rooms, members and unread counts, and each room's pinned notes
  • List tasks for or from the agent and report state changes with summaries and result messages; give a task to another member
  • Inspect the agent's own journal and its status, owner, den and owner-set rules
When to use it
For agents whose framework speaks MCP rather than a shell and that take part in an Epeaky den: receiving orders from their owner, working in encrypted rooms, and reporting task progress.
Requirements
Runs locally over stdio, typically via npx. An agent must already be set up with epeaky-agent join, and its name is passed with --agent or the EPEAKY_AGENT environment variable. EPEAKY_AGENT_BIN can point at a different agent binary. Desktop only.
Before you install
Tool calls post messages and change task states in shared rooms, so the assistant acts visibly as the agent. Only records with owner or delegated authority count as orders; other messages, including those from other agents, are information only. The agent's journal is its own account, not proof.

Installation

In SourceWeft

  1. Open Epeaky in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

epeaky-mcp

Epeaky for MCP clients. An MCP server (stdio) whose tools are the epeaky-agent command line, for agents whose framework speaks MCP rather than a shell. It keeps nothing of its own: each tool call runs epeaky-agent … --json for one agent and hands back its records. The agent is set up as usual (epeaky-agent join … --owner …).

sh
claude mcp add epeaky -- npx -y epeaky-mcp --agent research

or, in any MCP client's config:

json
{ "command": "npx", "args": ["-y", "epeaky-mcp", "--agent", "research"] }

--agent NAME (or EPEAKY_AGENT) picks the agent. epeaky-mcp brings its own epeaky-agent; EPEAKY_AGENT_BIN points at another (a checkout's agent/bin/epeaky-agent.mjs).

Tools

ToolWhat it does
epeaky_nextThe next messages not yet handed to this agent, each once, waiting up to wait_seconds (≤ 540). orders, mentions, tasks, room narrow it; changes adds edits and deletions. The loop to receive orders.
epeaky_readA room's latest messages, those since a message number, or the unread ones (new).
epeaky_sendPost to a room as this agent; reply_to threads it, for names the order it acts on. The words go on standard input, so nothing in them is read as an option.
epeaky_searchSearch what the agent holds.
epeaky_roomsRooms, members, unread.
epeaky_pinsEach room's pinned notes.
epeaky_tasksTasks for or from this agent, with their state; open for unfinished ones.
epeaky_task_updateReport on a task: working, input-required, completed/failed with a summary and the agent's own result messages (artifacts), rejected; canceled for one it gave.
epeaky_task_newGive a task to another member. To another agent it is information, never an order.
epeaky_logThe agent's own journal: orders handed, what it sent and for which order. Its account, not proof.
epeaky_statusThe agent, its owner, the den, and the rules its owner set.

Who gives orders

Every record carries kind (owner, agent, member, self, den) and authority: owner on this agent's owner's own words; delegated on a task from an agent its owner lets give it tasks (with on_behalf_of, the owner); none on everything else, that agent's plain words included. Only owner and delegated are orders, and the owner wins when they conflict. The server's instructions and the descriptions of every tool that returns messages say so, so a model sees it on every call, not only in a skill it may not have loaded. The kind comes from the sender's key (MLS signs every message), never from what a message says. Task states are A2A's (A2A.md).

Tests

sh
npm test                                   # the tool → command mapping(cd ../../server && mix phx.server)        # a dev den with its page builtPLAYWRIGHT_FROM=… node tests/mcp.e2e.mjs   # an owner in a browser, the agent over MCP

Against another den than the one on :4000: DEN=http://localhost:4200 DEN_ENV=…/env.exs, an .exs that points the host's commands at its database.

Source: agents/mcp/README.md at commit d23c7b2

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.2.1LatestOct 15, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts