AI Company

{
  "mcpServers": {
    "ai-company": {
      "command": "uvx",
      "args": [
        "ai-team-os"
      ]
    }
  }
}

claude mcp add --transport stdio ai-company -- uvx ai-team-os

{
  "mcpServers": {
    "ai-company": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "ai-team-os"
      ]
    }
  }
}

{
  "servers": {
    "ai-company": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "ai-team-os"
      ]
    }
  }
}

Config formats checked against the official docs on Oct 7, 2026: modelcontextprotocol.io (opens in a new tab), code.claude.com (opens in a new tab), cursor.com (opens in a new tab), code.visualstudio.com (opens in a new tab).

• team_status

  Get team details and member status

• team_list

  List all teams

• agent_update_status

  Update recorded Agent status

• agent_list

  List team members

• agent_template_list

  Get available Agent template list

• agent_template_recommend

  Recommend the best Agent template based on task description

• task_run

  Execute a task with full execution recording

• task_status

  Query task execution status

• task_create

  Create a new task (autostart supported; tasktype is accepted but retired — a no-op kept for backward compatibility)

• task_update

  Partial update of task fields with auto timestamps

• task_memo_add

  Add an execution memo to a task

• task_memo_read

  Read task history memos

• task_list_project

  List all tasks under a project

• meeting_create

  Create a structured meeting (8 templates, keyword auto-select)

• meeting_send_message

  Send a meeting message

• meeting_read_messages

  Read meeting records

• meeting_conclude

  Summarize meeting conclusions

• meeting_template_list

  Get available meeting template list

• meeting_list

  List all meetings

• meeting_update

  Update meeting metadata

• channel_send

  Send a message to a channel (team:/project:/global) with @mention support

• channel_read

  Read messages from a channel

• channel_wait

  Replay a scoped inbox, then wait for a peer message over WebSocket; read-only, no automatic ACK

• channel_mentions

  Get unread @mentions for an agent

• debate_start

  Start a structured 4-round debate (Advocate→Critic→Response→Judge)

• debate_code_review

  Start a code review debate session

• failure_analysis

  Frozen: record a failed task's root cause

• decision_log

  Log a decision to the cockpit timeline

• context_resolve

  Resolve current context and retrieve relevant background information

• memory_search

  Search team memory — recency-window recall within scope + pure-Python BM25 rerank (Chinese bigram, no embeddings)

• memory_add

  Write a direction-layer memory (preference/correction/design intent, 4 kinds; bucket quotas 1200/1500/300 chars, <=400…

• memory_invalidate

  Explicitly invalidate a direction-layer memory (by id or unique substring; invalidate, never delete — auditable);…

• memory_list

  List shared direction-layer entries, optionally filtered by kind

• memory_reconcile_candidates

  On-demand reconcile coarse pass (zero-LLM): BM25-paired candidate groups + direction-layer inventory + promotion…

• memory_reconcile_apply

  Apply agent-confirmed reconcile operations (merge / invalidate / score / promote); idempotent, size guardrails…

• unified_search

  Three-arm RRF search across memos / reports / tasks — BM25 full-text + knowledge-graph fanout + exact ID match

• link_query

  Query the cross-domain reference graph by node (what references this / what does this reference)

• link_trace

  Trace a reference chain from any OS ID (wf_id / commit / task uuid) with evidence snippets

• model_config_get

  Read observed Claude Code model names and its startup default

• model_config_set

  Set Claude Code's startup default with protected writes to its settings; does not control Codex

AI Team OS is a shared operating layer for Claude Code and Codex. Keep tasks, project memory, reports and team messages in one place, and follow work across sessions in one Dashboard. Each host keeps its native agent tools; the OS provides the durable record that makes their work understandable and reusable.

• agent-framework
• ai-agents
• claude
• claude-code
• dashboard
• fastapi
• mcp
• python
• react
• task-management