---
name: ipguru
description: Work with IPGuru through its MCP connector. Brainstorm with the crew, improve the Idea, make documents and images, and run Scenario Director. Use when the user mentions IPGuru.
compatibility: Needs the IPGuru MCP connector at https://mcp.ipguru.ai/mcp, signed in with the user's IPGuru account.
metadata:
  version: "1.1.0"
  publisher: "IPGuru.ai"
  homepage: "https://ipguru.ai/connect-skills.html"
---

# IPGuru

IPGuru is a guided AI for inventors. It takes an idea from a first thought to a developed
concept, with a crew of AI agents led by Nova and a panel of domain experts. The IPGuru
MCP connector gives you tools on the user's own IPGuru account. Everything you create or
change is saved to that account. The user can open the same work at https://chat.ipguru.ai.

## Before you start

1. Check that the IPGuru tools are available. The first tool is `open_session`.
   If there are no IPGuru tools, the connector is not set up. Help the user connect it with
   [references/connect.md](references/connect.md). Do not try to reach IPGuru any other way.
2. Call `open_session` first. It returns a session handle. Pass that handle to every other
   tool. It also returns the account summary and the user's scenarios.
3. Before a workflow, read the server's own guide for it: `playbook(mode)`, where mode is
   `harness`, `structural`, `pack`, `audit`, `studio`, `director` or `debate`. The playbook
   comes from the live server. If it disagrees with this skill, follow the playbook.
   `playbook` is a structure tool. If it is missing because the user added only a smaller
   connector, use that connector's MCP prompts, or suggest adding the main address.

## Words IPGuru uses

- **Scenario**: one idea being worked on. It has a transcript, an Idea record, sources,
  documents and Studio images.
- **Crew**: Nova (the lead) and the seated experts. They reply in turns.
- **Idea record**: the structured summary of the idea: decisions, will-nots, assumptions,
  next tests, gaps and a parking lot.
- **Card**: an option or recommendation the crew raised that waits for the user.
- **Scenario Director**: runs bounded loops of work towards an objective the user sets.

## Two ways to work

- **The crew does the thinking** (playbook `harness`). You carry the user's words to the
  crew and bring back their replies. Use this when the user wants IPGuru's brainstorm.
- **You do the thinking, IPGuru gives the structure** (playbook `structural`). You use the
  catalogues, the expander, the Idea record and the signals as tools, and you make the
  judgements. Use this when the user wants you to reason, with IPGuru's method as a frame.

Ask which one the user wants if it is not clear.

## Rules

1. **Ask before spending.** Tools tagged `scope:spend` run models or providers. They use the
   user's membership allowance, the same as in the app. Say so before the first metered step
   of a task, and before any step that starts a larger run, such as approving a Director
   plan. Never raise a limit the user set. Reads are free.
2. **Never invent the crew.** Report what the crew actually said, with speaker names and
   roles from the tool results. Never write a crew reply yourself.
3. **Report real states.** A queued job is not a finished job. A ready job is not a reviewed
   result. A high score is not proof that an objective was met. Say what the tool returned.
4. **Keep revisions current.** Many writes need the current `revision`. If a write answers
   with a conflict or a stale revision, read the record again, check what changed, then
   decide. Do not overwrite the user's newer work.
5. **Keep retry identities.** Keep `client_request_id`, `idempotency_key` and `command_id`
   values. After a lost reply, look the work up first (`turn_request`, `read_preparation`,
   the job or run status) and reuse the same identity. Do not start the work again.
6. **Long work is polled.** Every call returns within 45 seconds. A turn, job or run can take
   longer. Call the matching wait or status tool again while it reports `still_running`,
   and tell the user what is happening.
7. **Cards belong to the user.** After each turn, look at the open cards. A new message does
   not answer or close a card. Use `card_act` for the user's choice. Call
   `review_open_items` before you hand back to the user. Never close cards in bulk unless
   the user asks.
8. **Share the link.** Tools return an `app_url`. Give it to the user so they can open the
   same scenario in the app.
9. **Stay in scope.**
   - Prior-art searching belongs to IPGuru projects, not to these brainstorm tools. General
     research and reading sources are fine. An expert's opinion does not establish novelty.
   - Expert debate is a last resort. Use it only after discussion, questions, research and
     critique have failed to settle an important point.
   - Public publication is for IPGuru administrators only. It always needs the user's explicit
     approval of the exact frozen preview.
   - Confirm with the user before deleting anything.
10. **Pass errors through.** IPGuru's error messages are written for the user. Show them as
    they are. See [references/troubleshooting.md](references/troubleshooting.md).

## Common tasks

**Check the connection (read only).** `open_session`, then `list_scenarios`. Follow
`next_cursor` for more. Show each scenario's title, saved Idea score, freshness and
`app_url`. An empty list is a valid answer.

**Start a scenario the way the app's New Scenario wizard does.** Make a
`client_request_id` and keep it. Call `start_scenario` with the topic,
`preparation="wizard"`, `generate_title=true` and `mode="brainstorm"`, plus any theme,
frame, intent or experts the user chose. This can spend. Follow `read_preparation` until the
stages finish. Work continues on the server if the conversation ends.

**Run a crew turn.** `send_message` with the user's words, then `wait_for_turn` with the
returned `turn_id` until the status is not `still_running`. Show the new events with speaker
names. Then deal with the open cards (rule 7).

**Improve the Idea.** `read_idea` shows the record. `refresh_idea` (metered) rebuilds it
from the transcript. `score_idea` (free) saves the completeness score. `audit_signals`
(metered job) checks it; poll `signals_job`, then `read_signals`.

**Make a document.** `list_documents` shows the packs the account can make. `start_document`
(metered job), poll `document_run`, then `document_artifact` with `kind="markdown"`.
`document_export` builds Word or PDF where the account allows it.

**Make images in Studio.** Follow playbook `studio`. Read `read_studio` first, save the
description and art direction with their revisions, then commission runs. Look at every
image before approving or dismissing it.

**Run Scenario Director.** Follow playbook `director`. Read `read_director`, save the setup,
create a plan, show the user the exact plan, and approve it only when the user agrees.
Approval starts paid work.

More detail on each area:

- [references/brainstorm.md](references/brainstorm.md): sessions, scenarios, preparation,
  turns, cards, follow-ups, the crew and consultations.
- [references/structure.md](references/structure.md): the Idea record, signals, the
  expander, the pivot and the catalogues.
- [references/deliverables.md](references/deliverables.md): documents, references, web
  sources, the knowledge map and the Vault.
- [references/studio.md](references/studio.md): descriptions, art direction, shoot lists
  and image runs.
- [references/director.md](references/director.md): Scenario Director and publication.
- [references/troubleshooting.md](references/troubleshooting.md): errors and what to do.
- [references/connect.md](references/connect.md): how the user connects each assistant.

The full tool list is at https://ipguru.ai/mcp-reference.html.