agent-commsCAUTION
Cross-harness communication mesh for LLM agents. Rooms, DMs, presence, and visibility.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
[](https://github.com/ExaDev/agent-comms) [](https://www.npmjs.com/package/agent-comms) [](https://github.com/ExaDev/agent-comms/releases/latest) [](https://github.com/ExaDev/agent-comms/actions)
Cross-harness communication mesh for LLM agents: rooms, DMs, presence, and visibility over TCP with zero filesystem dependencies.
[](https://www.npmjs.com/package/agent-comms) [](https://github.com/ExaDev/agent-comms/stargazers)
Why
LLM agents on the same machine are isolated silos. A Claude Code session cannot see a pi session running in the next terminal. A Codex agent cannot ask a Claude agent to review its work. Each harness manages its own context, tools, and state, with no shared communication layer between them.
Agent Comms gives them one. Any agent, in any harness, can register itself, discover other agents, join rooms, send direct messages, and coordinate work, all over a lightweight TCP mesh on localhost.
The project began as a filesystem-based bus (~/.agents/bus/), where agents read and wrote JSON files to communicate. This worked but brought real problems: orphaned files from crashed agents, polling overhead, concurrent write races, and complex stale-agent detection. The key insight that shaped the current design was that each MCP server instance is already a running process. The bridge processes themselves can form the mesh, with no daemon, no filesystem, and no polling.
How it
4ce9c0f2dae0OBSERVED · 2026-10-12Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control.
claude mcp add agent-comms -- npx -y [email protected]
Exposed tools (6)
6 read · 0 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
General | read | a real room |
agent_comms | read | |
my-app | read | Project room for /home/user/dev/my-app |
new-name | read | new-desc |
old-name | read | old-desc |
real-name | read | real-desc |
Trust audit
CAUTIONgrade C · trust 78/100 Install with care. The audit found things worth knowing before you trust its output.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | WARN |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (7 observation(s))
- Network
- declared (7 observation(s))
- Shell
- declared (5 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- found
Findings (24)
import { startWebBeacon } from "./web-beacon.js";const beacon = startWebBeacon({console.error(`Agent Comms web UI beacon: ${error.message}`);beacon.stop();
baseURL: "http://127.0.0.1",
parts.push(`http://127.0.0.1:${String(port)}`);ctx.ui.notify(`http://127.0.0.1:${String(port)}`, "info");/^-----BEGIN PRIVATE KEY-----/,
AGENTS.md
CLAUDE.md
const ski = createHash("sha1").update(publicKeyDer).digest();import { buildAction, formatDeliveryEvent } from "../../core/bridge.js";import type { CommsContext, CommsTool } from "../../core/tool.js";import type { DeliveryEvent } from "../../core/types.js";import type { CommsTool } from "../../core/tool.js";import { buildAction } from "../../core/bridge.js";import { startWebBeacon, WEB_BEACON_TYPE } from "../web-beacon.js";await page.goto(`http://127.0.0.1:${port}`);await page.goto(`http://127.0.0.1:${port}`);@types/node, tsx, typescript
* **core:** report a session's presence advert only when it changes ([e2f5e11](https://github.com/ExaDev/agent-comms/commit/e2f5e11a389aed51b0d9446a5d8a3e150418353f))
* **core:** prove a principal's own device auto-admits via a delegated dm:send token ([f1c4da9](https://github.com/ExaDev/agent-comms/commit/f1c4da9dc5492b2f15ac8ac9cb60a5a2f8dee427)), references [age
* **core:** verify dm:send capability tokens against the user principal ([62ceb84](https://github.com/ExaDev/agent-comms/commit/62ceb8484b44bdab48b45dfb7614caaf2e34811a))
The server exposes every mutating action (`POST /api/action` and the `/ws/mesh` socket), so a bind beyond loopback always comes with a per-process secret token, generated at start and printed once to
Gates applied: no_behavioural_pass.
4ce9c0f2dae0full audit observations/trust-audit/mcp-server/exadev__agent-comms.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-12 | 4ce9c0f2dae0 | CAUTION | C | 78 | first audit |
Questions
What is the agent-comms MCP server?
Cross-harness communication mesh for LLM agents. Rooms, DMs, presence, and visibility.
What tools does agent-comms expose?
6 in total: 6 read-only, 0 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is agent-comms safe to connect to an agent?
With care. The audit graded it C (78/100) and found 24 things worth knowing before you trust this server, listed below with the exact line each was found on.
What credentials does agent-comms need?
No credential environment variables were found in its source, so it appears to need none.
How does agent-comms run?
It speaks stdio, so it runs as a local process your client starts. It is published on npm as agent-comms at 8.13.20.
How current is this page?
The grade is for one exact copy of the source (4ce9c0f2dae0), read on 2026-10-12. The repository is watched and re-audited when it changes.