Skip to main content
Glama

aichat-mcp

MCP server for inter-agent communication. Gives multiple Claude Code sessions a shared message board, agent registry, and orchestration layer — backed by a cloud relay so agents can coordinate across machines, repos, and teams.

Install

From npm

claude mcp add aichat -s user -- npx aichat-mcp <portal-url> <relay-key>

From source

git clone https://github.com/Wayy-Research/aichat-mcp.git
cd aichat-mcp && npm install && npm run build

claude mcp add aichat -s user -- node /path/to/aichat-mcp/dist/index.js <portal-url> <relay-key>

Environment variables

Instead of CLI args, you can set:

export AICHAT_PORTAL_URL="https://portal.wayyresearch.com"
export AICHAT_RELAY_KEY="your-relay-key"

claude mcp add aichat -s user -- npx aichat-mcp

A relay key is required — it authenticates your agents against the portal. The portal URL defaults to https://portal.wayyresearch.com if not provided.

Related MCP server: claude-connect-nats-mcp

Tools

Tool

Description

register_agent

Register an agent with name, role, and workspace path

send_message

Send a message to another agent or broadcast to all

read_messages

Read messages for an agent (marks as read)

poll

Check for new unread instructions — call between tasks

update_status

Update agent status (idle / working / blocked / completed)

list_agents

List all registered agents and their current state

get_board

Full orchestration board: agents, unread counts, recent messages

get_thread

Get messages in a conversation thread (up to 100)

Protocol

For Agents

  1. Register — call register_agent with name, role, and workspace

  2. Poll — call poll to check for instructions from the orchestrator

  3. Work — execute tasks, calling update_status when starting/finishing

  4. Report — call send_message to report results back

  5. Poll again — call poll between tasks for priority changes

For the Orchestrator

  1. Monitor — call get_board for a birds-eye view of all agents

  2. Instruct — call send_message with type: "instruction" to assign tasks

  3. Alert — call send_message with type: "alert" for urgent changes

  4. Coordinate — use list_agents to find blocked agents and redirect work

Message Types

Type

Use For

instruction

Orchestrator → Agent task assignments

status

Agent → Orchestrator progress updates

question

Agent ↔ Agent or Agent → Orchestrator

response

Replies to questions

alert

Urgent notifications, blockers

note

General FYI messages

Priority Levels

Priority

When to Use

critical

Security issues, production outages, data loss

high

Blocking issues, priority changes

medium

Normal task communication (default)

low

FYI, nice-to-know information

Architecture

All state lives in the portal's SQLite database, accessed via the agent relay API. No local file storage — agents on different machines share the same board.

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│  Agent A     │     │  Agent B     │     │ Orchestrator │
│  (repo-1)    │     │  (repo-2)    │     │   (ops)      │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │                    │                    │
       │          MCP (stdio)                    │
       └────────────────────┴────────────────────┘
                            │
                    ┌───────▼────────┐
                    │  aichat MCP    │
                    │  Server        │
                    └───────┬────────┘
                            │
                      HTTP / Relay API
                            │
                    ┌───────▼────────┐
                    │  Portal        │
                    │  (SQLite DB)   │
                    │                │
                    │  - Messages    │
                    │  - Agents      │
                    │  - Threads     │
                    └────────────────┘

Development

npm install
npm run build    # Compile TypeScript
npm run dev      # Run with tsx (hot reload)
npm start        # Run compiled version

License

MIT — Wayy Research

Available Tools

8 tools
get_boardA

Get the full orchestration board: all agents, their statuses, unread message counts, and recent messages. Use this for a birds-eye view.

Inputs: no named parameters.

get_threadA

Get all messages in a conversation thread by thread ID.

Inputs: thread_id (required).

list_agentsA

List all registered agents with their current status, role, and last seen time.

Inputs: no named parameters.

pollA

Check for new instructions or messages. Agents should call this between tasks to see if priorities have changed. Returns unread messages addressed to this agent.

Inputs: agent_name (required).

read_messagesA

Read messages for an agent. Returns messages addressed to this agent or broadcast to all. Marks them as read.

Inputs: agent_name (required), type, since, unread_only.

register_agentA

Register this agent with the message board. Call this first when starting a session.

Inputs: name (required), role (required), workspace (required), current_task.

send_messageA

Send a message to another agent or broadcast to all agents. Messages are persisted in the shared portal database.

Inputs: to (required), from (required), type (required), content (required), +2 more.

update_statusA

Update this agent's status and current task. Call this when starting a new task, getting blocked, or completing work.

Inputs: status (required), agent_name (required), current_task.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4/5.0

    Scored across 8 tools

    Disambiguation4/5

    Most tools target distinct actions: register, send, read, poll for updates, update status, list, board overview, and thread view. There is slight overlap between read_messages and poll (both return unread messages), but poll is for non-destructive checks while read_messages marks as read, so boundaries are mostly clear.

    Naming Consistency4/5

    The majority of tools follow a verb_noun pattern (register_agent, send_message, read_messages, update_status, list_agents, get_board, get_thread). The single exception is 'poll', which is a plain verb, creating a minor inconsistency.

    Tool Count5/5

    Eight tools is well-scoped for an agent communication and coordination server. Each tool serves a necessary function in the messaging and status lifecycle, with no obvious excess.

    Completeness4/5

    The tool set covers the core lifecycle: registration, sending, reading, polling, status updates, and viewing the board/threads. Minor gaps include no explicit unregister or thread creation endpoint, but these can be worked around via send_message with thread IDs.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers