WingmanPM MCP · Hosted · Preview

Your customer feedback, wired into your AI.

One hosted URL connects Claude, ChatGPT, or Cursor to your WingmanPM workspace — feedback, themes, priorities. Reads are instant. Writes wait for your go.

Read the MCP docs (opens in a new tab)

No API keys · OAuth sign-in · Nothing runs on your machine

MCP · hosted

Copy the server URL

Add it to your client

ChatGPT or Cursor? Paste the URL under Settings → Connectors.

Sign in when your browser opens

OAuth with a pairing code. No API keys, ever.

The round trip

What actually happens when your agent asks.

Four moves from question to confirmed write. Reads come straight back. Writes stop at the gate and wait for you.

Stage 01

Your agent asks.

A question leaves Claude, ChatGPT, or Cursor and lands on the hosted server. No API key, nothing running on your machine.

Stage 02

The server answers from your workspace.

Feedback, themes, prioritization scores — read straight from your data, up to ten items a call. Grounded, not guessed.

Stage 03

A write stops at the gate.

Anything that would change your workspace comes back as awaiting_confirmation with a one-time token. Nothing has happened yet.

Stage 04

You clear it. Once.

You see the exact tool and arguments, confirm, and the token dies in fifteen minutes. The agent polls for status — it never re-fires.

Thirteen tools. One rule.

Eight read, three write, two keep score. Every write stops and asks you first.

Reads

8 · instant

Ask anything: "what are customers angriest about this month?" — grounded in your workspace, not the model's imagination.

  • Products
  • Sources
  • Themes
  • Feedback search
  • Feedback detail
  • Prioritization scores
  • Task history

Writes

3 · confirmed

Long-running jobs your agent can start — after you approve the exact call.

  • Sync Intercom
  • Process a feedback file
  • Generate prioritization

Status

2

The handshake and the heartbeat.

  • Confirm task
  • Check task
Full tool reference
  • wingman_list_products
  • wingman_get_product
  • wingman_list_sources
  • wingman_list_themes
  • wingman_query_feedback
  • wingman_get_feedback
  • wingman_get_prioritization
  • wingman_list_tasks
  • wingman_sync_source
  • wingman_process_feedback
  • wingman_generate_prioritization
  • wingman_confirm_task
  • wingman_check_task

Every tool, with arguments and limits, in the MCP docs (opens in a new tab).

The differentiator

Writes wait for your go.

Most connectors hope an agent behaves. WingmanPM makes every write stop, show its work, and clear through a human before anything changes.

1

Agent proposes

The write comes back as awaiting_confirmation with a one-time token.

2

You see the exact call

Tool name and arguments, verbatim. No summaries, no surprises.

3

You confirm once

The token works exactly once and dies in 15 minutes.

4

Agent polls, never re-fires

Status checks until done. Ambiguous failure? It checks history instead of resubmitting.

No deletes·No publishing·No account admin·No roadmap edits—those tools don't exist.

Workspace data is treated as untrusted input, never as instructions.

The skill

Access is half the job. Technique is the other half.

The WingmanPM skill teaches your agent how a senior PM actually works the data — read first, write last, show your evidence.

  • Feedback triage
  • Prioritization runs
  • Theme review
  • Trend scans
  • Grounded PRDs
  • User stories & epics
  • Stakeholder updates

Plain Markdown files. Auditable in any editor. Nothing executes on install.

Works where your agent works

Claude Desktop·Claude Code·Claude.ai·ChatGPT·Cursor·any MCP client

Hosted MCP · Preview

Put a wingman in the loop.

Hosted MCP in preview. Connect in under a minute; writes always clear through you.

Read the docs (opens in a new tab)