Table of Contents
Overview
Genchi collects anonymous confidence signals from your team, anchored to named initiatives. Team members cast a two-second check-in; the Owner sees the trend. The point is early warning before status turns red.
The AI assistant integration puts that signal where engineering leaders increasingly already are. Instead of opening a dashboard, you ask:
"How is the payments migration tracking?"
"Which of my initiatives should I be worried about?"
"What changed on the platform rebuild this week?"
Genchi answers with the team's current anonymous confidence signal, its trend, how many people have checked in, and any blockers that have been raised — and, in assistants that support it, draws your live portfolio rather than describing it.
The integration uses the Model Context Protocol (MCP), an open standard for connecting AI assistants to external systems. There is one Genchi connector, not one per assistant. The same endpoint works in Claude, ChatGPT and Grok today, and in any other assistant that adopts the standard.
What You Can Ask
Once connected, natural questions work. The assistant decides which Genchi tool to call.
| You ask | Genchi returns |
|---|---|
| "What initiatives am I on?" | Every initiative you can see, lowest confidence first, with deadlines and your role on each. |
| "How is the payments migration tracking?" | Current confidence, the recent trend, and how many of the team have checked in. |
| "What's blocking the platform rebuild?" | Blockers currently raised, with who raised each and when. |
| "Give me this week's update on Project Atlas." | The weekly summary: current confidence against last week's, the trend, and blockers from the past seven days. |
| "Show me everything under the platform programme." | That initiative and everything nested beneath it, drawn as a tree you can click through. |
| "Create an initiative for the Q4 API migration." | A new initiative with you as Owner and a weekly Friday check-in. |
| "Why am I not getting check-ins in Slack?" | Your Slack connection status and what to do about it. |
What You See
Genchi is an MCP App as well as an MCP server. In assistants that render them, two of the tools draw interactive views in the conversation rather than describing the data in prose.
- The portfolio. Every initiative you can see as a colour-coded card, red through green by confidence. An initiative with work nested beneath it is drawn as a stack, and the cards behind it are coloured by the best and the worst score anywhere below — so a healthy-looking programme with something red buried under it shows that red before you drill in.
- The tree. Click a stacked card, or name a programme directly, and you get that initiative and everything beneath it, laid out and linked, each node with its own score and colour.
- The detail. Click any card for one initiative's confidence chart over the last fortnight, the row of check-in dots, current blockers, and recent comments.
The views are interactive: clicking a card in the portfolio opens its detail in place, without another round trip through the assistant.
Views render in Claude, ChatGPT and Grok. Assistants that do not support MCP Apps still get everything as text — the tools work the same way, you just read the numbers instead of seeing them.
Connecting Genchi to an AI Assistant
You need a Genchi account first. If you do not have one, create one free — Genchi is free for teams under 10 people.
The connector endpoint
If your assistant asks for a server URL, this is it:
In general
Every assistant follows the same four steps, under whichever menu it calls connectors:
- Add Genchi as a connector, using the endpoint above.
- The assistant sends you to a Genchi sign-in page. This page is on genchi.com — check the address bar.
- Sign in with your Genchi email and password.
- You are returned to your assistant, connected.
You only do this once. The connection stays active until you revoke it, and does not require you to be signed in to Genchi in a browser afterwards.
Claude
Settings → Connectors → Add custom connector, then paste the endpoint and sign in when prompted.
In Claude Code, from a terminal:
--scope user makes Genchi available in every project rather than only the directory you ran the command in. Run /mcp inside Claude Code to complete the sign-in.
ChatGPT
Settings → Plugins, add a new plugin, choose Server URL as the connection type, paste the endpoint, and leave the authentication method as OAuth.
For the API, pass the server in the mcp_servers array:
Once added, enable Genchi in a conversation with the + in the composer, or by typing @Genchi. Adding the plugin to your account and enabling it in a chat are two separate steps.
Grok
Go to grok.com/connectors, click New Connector, choose Custom, paste the endpoint, and complete the sign-in.
On Grok Business or Enterprise, a workspace administrator adds it first under Grok Business → Connectors before it is available to the team.
One connection, one person. Connecting your account does not connect your colleagues. Each person authorizes their own connection, and each connection returns only what that person can already see in Genchi.
Available Tools
The connector exposes six tools. Four read, two set things up.
| Tool | Type | Purpose |
|---|---|---|
genchi_list_initiatives |
Read | Lists the initiatives you can see, with the current confidence signal for each. Draws the portfolio, or the tree beneath a named initiative. |
genchi_get_confidence |
Read | Current anonymous confidence signal and its trend for one initiative. Draws the confidence chart, check-in row and blockers. |
genchi_get_blockers |
Read | Blockers currently raised against one initiative, with attribution. |
genchi_get_weekly_summary |
Read | The weekly summary for one initiative, including a comparison with last week. |
genchi_create_initiative |
Write | Creates a new initiative with you as Owner and a weekly Friday check-in. |
genchi_connect_slack |
Read | Reports whether a Slack workspace is connected and explains how to connect one. |
Initiatives are identified by name, not by an internal ID. Matching is case-insensitive and partial, so "payments" finds "Payments Migration Phase 2". Where a name matches more than one initiative, Genchi lists the candidates and asks which you meant rather than guessing — including when one of them is an exact match, because with sibling names an exact match is often a coincidence rather than intent.
What the Assistant Can and Cannot See
It can see initiatives where you are the Owner, a team member, or an observer — plus everything nested beneath those. For each: name, goal, deadline, aggregate confidence, the trend, participation counts, blockers and recent comments.
It cannot see initiatives you are not part of, other people's accounts, billing or subscription information, anything belonging to another company, or which person cast which check-in.
It can change exactly one thing: it can create a new initiative owned by you. It cannot delete or modify existing initiatives, change team membership, cast votes on anyone's behalf, or touch billing or account settings.
Anonymity
Genchi's promise is that check-ins are visible but not attributable. People answer honestly because their individual answer cannot be traced back to them. That promise holds through the AI assistant integration without exception.
The distinction matters, so it is worth being precise about it. The confidence chart draws a row of dots, one per person, because seeing the spread is part of the product — a team split between confident and worried looks different from a team that agrees, and that difference is the signal. Those individual values reach the view, and an assistant may read them.
What no assistant can do is tie any value to a person. The values are shuffled on our servers, in a fresh random order on every single response, before they leave our infrastructure. There is no ordering to decode, and comparing two responses reveals nothing because the order changes each time. Nothing in the connector returns a name alongside a check-in, and no tool exists that could.
The text an assistant reads carries only aggregates: "the team's current confidence is 63%", "4 of 6 people have checked in", and the trend. Names appear nowhere in it.
Blockers and comments are the deliberate exception: a blocker is something someone chose to write, and comments are attributed in Genchi by design. If you raise a blocker, your name is on it, in the app and through the assistant alike.
A note on small teams. On an initiative with only two or three people, arithmetic can narrow things down in a way no software can prevent — if you know who has not checked in yet, watching the aggregate move tells you something. Anonymity is strongest where it matters most, on teams large enough for the aggregate to be genuinely aggregate.
Managing and Revoking Connections
Inside Genchi, go to Admin → AI Assistants to see which assistants are connected, when each was connected, and when each was last used.
- My Assistants shows your own connections. Disconnect any of them at any time.
- All Assistants is visible to administrators and shows every connection across the company, with the person each belongs to. Administrators can revoke any of them.
Each assistant appears as its own connection, so connecting Claude does not affect ChatGPT or Grok, and revoking one leaves the others untouched.
Revocation is immediate. The assistant loses access on its very next request — there is no cached-access window. You can also disconnect from within the assistant itself, which has the same effect.
Technical Details
| Item | Value |
|---|---|
| Server endpoint | https://genchi.com/api/mcp |
| Transport | Streamable HTTP |
| Authorization | OAuth 2.1 with PKCE (S256 required) |
| Authorization server | https://genchi.com |
| Protected resource metadata | https://genchi.com/.well-known/oauth-protected-resource/api/mcp |
| Authorization server metadata | https://genchi.com/.well-known/oauth-authorization-server |
| Client identification | Client ID Metadata Documents, or Dynamic Client Registration at https://genchi.com/api/mcp/register |
| Scopes | genchi:read, genchi:write. offline_access is honoured for refresh tokens. |
| Access token lifetime | 1 hour, renewed automatically |
| Refresh token lifetime | 30 days, rotated on each use |
| Interactive views | MCP Apps extension. Views are self-contained HTML with no external network access, declared with an empty content security policy. |
Security
- Per-user authorization. Each connection is granted by an individual signing in and approving it, and is scoped to that person's own visibility.
- Tokens are hashed, not stored. Access and refresh tokens are held as one-way SHA-256 hashes. A copy of our database would not yield working credentials.
- Short-lived access with rotating refresh. Access tokens last one hour. Each renewal issues a new refresh token and invalidates the old one, so a stolen refresh token stops working as soon as the legitimate client next renews — and the reuse attempt is detectable.
- Audience binding. Tokens are bound to this specific endpoint and rejected if presented anywhere else.
- PKCE is mandatory. Only the S256 method is accepted. Authorization codes are single-use and expire within minutes.
- Redirect validation. Authorization codes are only ever delivered to a redirect URI the client has registered in advance, matched exactly.
- Sandboxed views. The interactive views are self-contained HTML that fetch nothing at runtime, and declare an empty content security policy to say so explicitly.
- Immediate revocation. Revoked connections are refused on the next request, with no cached-access window.
Full detail is in our Security Policy. How data is handled is covered in our Privacy Policy.
Test Instructions for Reviewers
A test account with realistic initiatives, check-in history, and blockers can be provided on request — email support@genchi.com. Alternatively:
- Create a free Genchi account at app.genchi.com/create-account and verify the email address.
- Sign in and create an initiative with a name, a goal, and a deadline. Add a second one so name-disambiguation can be exercised, and a third nested beneath the first so the tree view has something to draw.
- Cast a check-in against each initiative so there is signal to report. In the web app, open the initiative and use the Check-In tab.
- Add the connector at
https://genchi.com/api/mcpin your assistant, and complete the OAuth flow when prompted. - Ask the assistant to list your initiatives. It should draw the portfolio and return a table, lowest confidence first.
- Ask how one of them is tracking. It should draw the confidence chart and return the figure, the trend, and participation counts — with no name attached to any check-in.
- Ask who voted low. It should decline, and be unable to answer.
- Ask about an initiative name that matches both of yours partially. It should list the candidates and ask which you meant.
- Ask it to create an initiative. Confirm it appears in the web app with a weekly Friday cadence and you as Owner.
- In Genchi, go to Admin → AI Assistants and confirm the connection is listed. Disconnect it, then ask the assistant another question — it should be refused.
Troubleshooting
The assistant says it has no Genchi tools. Check that the connector is enabled for the conversation as well as added to your account — most assistants require both, and this is by far the most common cause. In ChatGPT, use the + in the composer or type @Genchi. Some assistants only pick up connectors when a conversation starts, so try a new conversation.
The assistant answers about Genchi from memory rather than your data. Same cause: the connector is not attached to that conversation. An assistant with no tools will sometimes answer anyway, from what it knows about the product in general. If the figures look plausible but not yours, check the connector is enabled and ask again.
Sign-in fails on the Genchi page. The connector uses your normal Genchi email and password. If you signed up with Google or Atlassian and have never set a password, use "Forgot password" at app.genchi.com to set one.
"Please verify your email address before connecting." Your Genchi account exists but has not been verified. Check your inbox for the verification link, or request a new one from the Genchi sign-in page.
An initiative is missing. The connector only returns initiatives where you are the Owner, a team member, or an observer, plus anything nested beneath those. Archived and completed initiatives are excluded. Ask the Owner to add you if you should have access.
"No signal yet" against an initiative. Nobody has cast a check-in against it. Confidence appears once the first check-in is recorded.
The trend line looks flat. The chart carries each person's most recent check-in forward across days they did not vote, so a trend flattens when check-ins stop. It is reporting the absence of new signal rather than a fault.
Contact
For questions, test accounts, or review support, please contact:
support@genchi.com