1. Conduid
  2. Developer Tools
  3. Agentishare
MCP server · Developer Tools

Agentishare

uh um uh yeah

Unclaimed MIT last commit 7 months ago devtools
59Fair

Scored 6 days ago · breakdown

About Agentishare

Agentishare is an MCP server published by JTP75 in the Developer Tools category: uh um uh yeah.

The repository has 1 stars and 0 forks, with the last commit 7 months ago. Six months or more without a commit doesn't mean the server is broken, but check the open issues before depending on it in production.

Install

From source
git clone https://github.com/JTP75/agentishare

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Tools (7)

agent_hub_list_agents()List agents currently in your team

Parameters not published.

agent_hub_receive()Fetch and clear your inbound message buffer

Parameters not published.

agent_hub_reset()Disconnect and erase stored credentials for this workspace; restart required to re-setup

Parameters not published.

agent_hub_send(to, type, content)Send to a named agent or `"broadcast"` to all

Parameters not published.

agent_hub_setup_create(agentName)Create a new team; returns an `apiKey` to share with collaborators

Parameters not published.

agent_hub_setup_join(apiKey, agentName)Join an existing team using a shared key

Parameters not published.

agent_hub_whoami()Return your agent name, team ID, and connection info

Parameters not published.

Ask AI

✦

Ask AI about Agentishare

Powered by Claude · Grounded in docs

I know everything about Agentishare. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

Static checks derived from the repository 5 days ago. Heuristics on the source, not an audit.

  • ✓README presentThe repository ships documentation.
  • ✓License declaredUsable under a stated license.
  • ✓Tests presentA test suite exists in the repository.
  • ✓Dependencies pinnedA lockfile pins dependency versions.
  • ✓No dynamic code executionSource scan found no eval, dynamic Function, or shell execution.
  • !Scoped permissionsDoesn't declare a permission scope. Assume it can do anything its process can.

README

Agent Hub

Free, open-source MCP server for real-time context sharing between AI coding tools.

Problem: Developer A (Claude Code) builds an API. Developer B (Cursor) needs those specs to build the frontend. Today: copy/paste through Slack. With Agent Hub: "Send my API endpoints to David's Cursor instance."


Transports

Agent Hub supports two transport modes:

Nostr (default) — agents communicate peer-to-peer via a public Nostr relay. No hub server required. Team membership is a shared secret; agent identity is a secp256k1 keypair generated on first use.

Claude Code ──[stdio]── mcp-client ──[WebSocket]──┐
                                                   ├── Nostr relay (e.g. wss://nos.lol)
Cursor      ──[stdio]── mcp-client ──[WebSocket]──┘

Hub — agents connect through a central Express server. Supports remote MCP (no local install needed for Claude Code) and provides a hosted public instance.

Claude Code ──[stdio]── mcp-client ──[HTTP+SSE]──┐
                                                  ├── hub-server
Cursor      ──[stdio]── mcp-client ──[HTTP+SSE]──┘

Setup

Option A — Nostr transport (default)

No hub server required. Add the client and call a setup tool:

claude mcp add agent-hub --scope user -- npx @agent-share/mcp-client

Then from within your AI session, call agent_hub_setup_create or agent_hub_setup_join. Your generated private key and team ID are saved to ~/.config/agent-hub/config.json — future sessions connect automatically.

The default relay is wss://nos.lol. Override with HUB_URL=wss://your.relay.

Known limitation: public Nostr relays may not persist custom event kinds. Messages sent to a disconnected agent can be lost if the relay drops them before the agent reconnects.

Option B — Hub transport

Use the hosted instance (no setup required):

The public hub is live at https://agent-hub-wild-glade-1248.fly.dev.

Or run your own locally:

npm install
NODE_ENV=development npm run dev

Or deploy to Fly.io:

fly volumes create agent_hub_data --size 1 --region iad --config packages/hub-server/fly.toml
fly deploy --config packages/hub-server/fly.toml

State persists to a Fly volume at /data. Verify after deploy:

fly ssh console --config packages/hub-server/fly.toml -C "cat /data/agent-hub.json"

2. Connect your agent

No credentials required at startup. Add the client, then run a setup tool from within the session:

claude mcp add agent-hub --scope user -- npx @agent-share/mcp-client

Then from within your AI session, call a setup tool to create or join a team (see Tools below).

To use the hub transport explicitly, set TRANSPORT=hub in your MCP client env or pass HUB_URL to point at a custom instance.


Tools

Available to your AI agent after connecting:

Tool Description
agent_hub_setup_create(agentName) Create a new team; returns an apiKey to share with collaborators
agent_hub_setup_join(apiKey, agentName) Join an existing team using a shared key
agent_hub_send(to, type, content) Send to a named agent or "broadcast" to all
agent_hub_list_agents() List agents currently in your team
agent_hub_receive() Fetch and clear your inbound message buffer
agent_hub_whoami() Return your agent name, team ID, and connection info
agent_hub_reset() Disconnect and erase stored credentials for this workspace; restart required to re-setup

Message types: api_spec, file_change, decision, todo, question

Example:

"Send my current API endpoint definitions to bob as an api_spec message."

Credentials are saved per-workspace to ~/.config/agent-hub/config.json. Each project directory gets its own team entry. Env vars (TEAM_API_KEY, AGENT_NAME, HUB_URL, TRANSPORT) take priority over the config file and are useful for CI/scripted use.


Configuration

All values live in config/*.props — nothing is hardcoded in source.

Section Key Default Description
server port 3000 Hub listen port
rate_limit max_requests 100 Requests per minute
sse max_message_buffer_size 100 Max buffered messages per agent
team max_agents_per_team 20 Max agents per team
nostr relay_url wss://nos.lol Default Nostr relay
nostr heartbeat_ms 60000 Presence re-publish interval
nostr presence_window_s 90 Subscription lookback window
hub default_url (hosted instance) Default hub URL for the stdio client

Implementation

  • packages/hub-server — Express + @modelcontextprotocol/sdk. Each SSE connection spawns its own McpServer instance so tools are scoped to the authenticated agent. Persistence uses a pluggable IStore interface: MemoryStore (dev), VolumeStore (Fly.io, active in prod), RedisStore (multi-machine). Swap by changing one import in store/index.ts.
  • packages/mcp-client — Stdio MCP server. Transport is abstracted behind ITransport; HubClient (HTTP/SSE) and NostrClient (WebSocket + nostr-tools) are the two implementations. Starts in bootstrap mode if no credentials are found.
  • Auth (hub) — API keys are 64-char hex, SHA-256 hashed at rest. Agents pass via ?api_key= or Authorization: Bearer.
  • Auth (Nostr) — Permissionless. Team membership is a shared 32-char hex string used as a Nostr tag filter. Agent identity is a secp256k1 keypair generated on first use and saved to the local config.
  • Tests — 36 unit tests for mcp-client (npm test -w packages/mcp-client), integration tests against a live relay (npm run test:integration -w packages/mcp-client).

License

MIT

README fetched from the repository 5 days ago. The original is authoritative.

Questions

About Agentishare

How do I install Agentishare?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is Agentishare safe to use with an AI agent?

Its trust score is 59 out of 100 (fair). It passes 5 of the 6 static security checks run so far; the failures are listed above. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Agentishare still maintained?

The last commit was 7 months ago. That's long enough that you should check whether the maintainer is responding to issues before depending on it.