AIOProductOS MCP · server

One MCP for your whole product. Manage the board, read your product brain + analytics, and run the support inbox from your AI host — Claude Code, Cursor, Codex, or any MCP client — all reasoning on your own model subscription (BYOK; AIOProductOS never charges for the agent's compute). Thin and secret-free: the only credential is your personal access token, which you paste into your host's config.

1. Get a token

In AIOProductOS → Settings → Tokens & Agents → Generate token. Copy it (shown once) — it looks like pat_live_…. The token acts as you, with your access.

2. Connect

Claude Code

claude mcp add productos -e PRODUCTOS_TOKEN=pat_live_… -- npx -y @aioproductoscom/mcp

Cursor / Codex — .mcp.json

{
  "mcpServers": {
    "productos": {
      "command": "npx",
      "args": ["-y", "@aioproductoscom/mcp"],
      "env": { "PRODUCTOS_TOKEN": "pat_live_…" }
    }
  }
}

PRODUCTOS_URL defaults to https://platform.aioproductos.com; set it to point at a self-hosted platform.

Prefer no local install? Use the hosted remote

AIOProductOS is also a hosted remote MCP (OAuth 2.1 sign-in, no token pasting, the same 71 tools) — add the custom connector URL https://platform.aioproductos.com/api/mcp in Claude.ai, Claude Code, or any remote-capable MCP host. Listed on the official MCP Registry as com.aioproductos/mcp and verified on Glama.

Tools — 71

It reads and writes across the whole strategy-to-delivery ladder, so an assistant can run product work end to end instead of reporting on it. Identical tool set to the hosted remote connector.

Orientation — whoami (org + member), get_pm_playbook (how to operate), pm_meta (lists / statuses / members / features — resolve names → ids)

Grounding — get_product_brain (revenue, accounts, analytics, features, signals, open work in one read), get_weekly_signal_memo, get_customer_360 (money + people + voice for one account), get_codebase_map

Discovery — ideas — list_ideas, create_idea, update_idea, vote_idea, promote_idea (idea → feature)

Strategy — initiatives & OKRs — list_initiatives, create_initiative, update_initiative, list_objectives, create_objective, update_objective, update_key_result

Roadmap — list_features, create_feature, update_feature, list_releases, create_release, update_release, get_roadmap_drift (planned vs shipped, scored)

Delivery — board & sprints — list_tasks, get_task, create_task, update_task, delete_task, comment_on_task, list_sprints, create_sprint, update_sprint

Docs & decisions — list_pages, get_page, create_page, update_page, list_decisions, create_decision, update_decision

Experiments — list_experiments, create_experiment, update_experiment

Voice of customer — list_insights, capture_insight

Analytics — analyze_nps, analyze_nrr (both revenue-weighted), analyze_funnel, get_retention, analyze_paths

Support inbox — list_conversations, get_conversation, reply_to_conversation (visible in the visitor's chat widget), add_note (internal only), resolve_conversation

Scheduling — list_bookings, cancel_booking, reschedule_booking

Team comms — list_channels, read_channel, post_to_channel, reply_in_channel

Identity graph — get_device_candidates, merge_end_users, unmerge_end_users (reversible 30 days), list_identity_merges

AI artifacts — review_artifact (critic vs your PM bar), list_artifact_versions, revert_to_version

Every tool declares read-only / write / destructive hints, so a host can gate them: 34 read, 37 write, 1 destructive. Writes are scoped to your org and respect your role's permissions — the token can never reach another tenant.

Three groups of writes are worth knowing about before you connect, because they are visible to someone other than you:

• Customer-facing — reply_to_conversation, add_note and resolve_conversation act on your support inbox; a visitor sees the reply.
• Team-visible — post_to_channel and reply_in_channel post as you.
• Guest-visible — cancel_booking and reschedule_booking change a booked meeting and notify the guest.

delete_task is the only destructive tool. It is permanent, cascades to subtasks, and has no undo — so it additionally requires confirm: "DELETE". A task id alone can never trigger it; an assistant has to ask you first.

For an assignable coding teammate (Backend / Frontend / Mobile / QA that opens PRs in your repo), install the separate @aioproductoscom/mcp-agent.

Interactive apps — 3

Three tools render a real interface inside the conversation (MCP Apps, SEP-1865) instead of a wall of JSON. They are not screenshots: each one talks back to this same MCP over the app bridge.

| Tool | App | What you can do in it | |---|---|---| | list_tasks | Board | Tasks grouped by status; change a status and it writes through update_task | | get_weekly_signal_memo | Weekly signal | This week's themes with evidence counts; turn any theme into a linked task | | analyze_funnel | Funnel | Conversion per step with the live MRR standing behind each one |

Your host needs MCP Apps support to render them (Claude, ChatGPT, VS Code, Goose and others do). Everywhere else the tools return exactly the text they always did — the app is an enhancement, never the only copy of the answer.

Links

• Website: aioproductos.com
• Platform: platform.aioproductos.com
• MCP docs & host picker: aioproductos.com/mcp
• Claude Code plugin (one install = connector + /triage command): github.com/AIOProductOS/claude-plugin
• Support: [email protected] — a human founder answers every email.

License

MIT. The package is free; it connects to a paid AIOProductOS workspace — every plan has a 14-day onboarding runway and a 30-day money-back guarantee.