Using HQ from ChatGPT and Claude.ai

HQ has a connector (called HQ MCP) that plugs your HQ into ChatGPT and Claude.ai. Once it's connected, the chat assistant you already use can see the same governed HQ context your local sessions use: company knowledge, project records, policies, and vault files, with the same permissions you have everywhere else in HQ. Nothing needs to be installed on your computer; the connector talks to HQ's cloud directly.

One thing to know up front: the connector is currently in beta, on HQ's side and on the ChatGPT and Claude sides too. Setup screens, labels, and permissions may shift, and you should review write actions before approving them.

What you can do from chat#

Once your Champion has set the connector up (and you've connected your own account), you can ask ChatGPT or Claude to:

• Browse and read your files. List the companies your account can access, browse readable folders in the company vault, and open files.
• Search company knowledge. Search across the HQ content your account is allowed to see, and read knowledge docs, project records, project status, and policies.
• Capture things back into HQ. Save notes, journal entries, project updates, and knowledge captures into your HQ's writable areas, so the thought survives the chat.
• Share files. Share a company file or folder either as a short-lived link (a browser page where you pick the recipients and their read or write access) or as a direct grant to one teammate, a group, or the whole company. Read access is the default; the same rules from Sharing Your Work apply.
• Publish small static pages. Deploy a lightweight page (a readout, a preview, a handoff artifact) straight from chat, with public, password-gated, or company-only access. Bigger builds and real apps still go through the local /deploy workflow.
• Use and propose team skills. Fetch your team's approved workflow skills as step-by-step guidance, and draft new skills for review. Chat can never write a live skill directly: drafts go to an owner or designated reviewer, and only approved drafts are promoted.

A note on secrets: chat can see secret names and what a project needs, and it can generate a secure one-time link for a human to supply a value, exactly as described in the Secrets Guide. It cannot see or return secret values.

What stays safe#

The connector is built so the safety rules you rely on in HQ hold in chat too:

• Secret values never appear in the transcript. Chat gets names, schemas, and submission links only; values can't be returned to the model.
• Company walls hold. Every request checks which companies your account belongs to first. One company cannot read or write another company's content through chat.
• Files you create are yours until you share them. A file written from chat is owned by you, its creator, just like a file synced from your laptop. Teammates get access only when you explicitly share it.
• Sensitive folders are never readable. The connector excludes private operational areas (settings, secrets, and workers paths) from listings and reads entirely.
• You approve every write and share. Your chat client shows an approval prompt before a write or share action runs. Treat that prompt as your final human review: read what's about to happen before you say yes.

Setting it up (Champions)#

Everything below is one-time setup. Have these values ready; they're the same for every client:

ChatGPT

1. Ask a ChatGPT workspace admin or owner to enable developer mode for custom MCP apps (under Workspace Settings, in Permissions & Roles / Connected Data). Some plans also have a per-user toggle under Settings, then Apps, then Advanced Settings.
2. Create a custom app (from Workspace Settings, then Apps, then Create, or from your own Settings if your workspace allows it).
3. Enter the MCP server URL, choose OAuth, and provide the Client ID above if asked. ChatGPT's OAuth scopes are openid, email, profile, and hq-mcp/use.
4. Click Scan Tools. When the HQ authorization flow opens, sign in with your HQ identity and approve the access.
5. Click Create. The app appears with a developer or draft label until it's published for the workspace.
6. Each user then selects the HQ app from the tools menu in a new chat, or refers to it by name in a prompt.

Claude.ai on a Team or Enterprise plan

1. Ask an owner or primary owner to open Organization settings, then Connectors.
2. Click Add, choose Custom, then Web, and enter the MCP server URL. Open Advanced settings only if Claude asks for OAuth client details, and enter the Client ID there.
3. Click Add.
4. Each user then opens Customize, then Connectors, finds the HQ connector, clicks Connect, and completes the HQ sign-in.

Claude.ai on a Pro or Max plan

1. Open Customize, then Connectors.
2. Click +, then Add custom connector, and enter the MCP server URL (Client ID under Advanced settings only if asked).
3. Click Add, then Connect, and complete the HQ sign-in.

In either Claude setup, users enable the connector per conversation from the + menu in the chat composer, under Connectors.

Troubleshooting#

• The tool list looks stale (a capability this page mentions is missing, or tools look outdated): the chat client cached an old tool list. In Claude, remove and re-add the connector so it rescans. In ChatGPT, refresh the app's actions or recreate the draft app.
• Tools appear but every HQ action fails: the connection to your HQ identity has gone stale. Disconnect and reconnect the connector so it signs in again.

Common questions#

If you get stuck#

Start with the two troubleshooting fixes above; they cover most beta rough edges. If the connector still misbehaves, tell your Champion (they own the workspace-level setup) and report it with /hq-bug from a regular HQ session. For everything unrelated to chat, head to The "I'm Stuck" Guide or the FAQ.