What it is
The AI Connector is an address you add to an AI assistant so it can use TagPipes while you chat. Ask "what should I pay attention to this week?" or "break last month's orders down by source", and the assistant gets the answer from Lens instead of guessing. It uses the open Model Context Protocol (MCP).
The page is in beta. Open it from the Lens sidebar under Configure → AI Connector.
| Assistant | Status |
|---|---|
| Claude (claude.ai, Claude Desktop) | Tested |
| Gemini CLI | Not tested yet |
| Other MCP clients that can add a remote server and send a request header | Not tested yet |
| ChatGPT | Not supported yet: its connectors cannot send an access key and need a sign-in flow TagPipes does not offer yet |
Before you start
The Readiness panel on the right must show both items as ready:
| Item | Needs |
|---|---|
| Default analysis area | Set. The connector answers every question from the property's default analysis area, so nothing can be analyzed without one. Choose or create one in Lens Models. See Semantic Models. |
| Your access key | Ready. If it says Not created, open your profile, go to the API tab, click Generate, then Save, and come back. |
Connect Claude
- On the AI Connector page, copy the Connector URL and the Suggested connector name with their Copy buttons.
- Under 2. Your access key, click Copy. Show key reveals it if you need to check it.
- In claude.ai, open Customize, then Connectors, and choose Add custom connector.
- Paste the suggested name as the name and the connector URL as the URL.
- Add a request header named
x-api-keywith your access key as its value. - Save. The connector should show as connected.
Claude Desktop uses the same connectors as claude.ai once you are signed in. For Gemini CLI, the page gives a command to copy; replace YOUR_ACCESS_KEY with your key before running it. For other clients, add a remote MCP server (Streamable HTTP) with the connector URL and the same x-api-key header.
Keep questions on TagPipes
If your assistant has other tools connected, such as email, calendar or ticketing, a broad question can go to the wrong one. Give this client its own space:
- In Claude, create a Project for this client and turn on only this connector. In other assistants, use the equivalent: a workspace, custom instructions or a system prompt.
- Copy the Project instructions from section 4 of the page and paste them there.
Check it worked
Start a new chat (a chat that was already open keeps the old list of tools) and ask, for example:
- What should I be paying attention to this week?
- Investigate the biggest change.
- Can I trust this number?
- What does a company term mean here?
What the assistant can and cannot do
| It can | It cannot |
|---|---|
| Ask Lens questions, with the same analysis areas and definitions as Analyze | Write its own queries or formulas, or read raw data tables |
| Read the latest Insights findings | Change any setting, table, analysis area or connection |
| Explain how a measure is defined | Create, edit, reorder or delete boards or tiles |
| Check whether a number can be trusted, and look up events TagPipes recorded around a change | Publish a report |
| Look up your company's business terms and your brand palette | Use another person's access |
| Add an analysis to an existing board, when you ask |
Adding an analysis to a board is the only change it can make, and only to a board that already exists.
TagPipes asks the assistant to label every statement: Verified by Lens, Calculated by Lens, Derived from Lens, Recorded by TagPipes, or AI inference. Only the first four come from TagPipes. AI inference is the assistant's own reasoning, so treat it as a suggestion. Claude follows these labels; other assistants have not been checked yet.
Scope and access
| Connector URL | Answers from |
|---|---|
| The property Connector URL | This property only |
| The Company connector URL, under All properties of your company | Every property of the company you can access. The assistant asks which one when it is not clear. |
The access key is your personal TagPipes API token, the same one on your profile. The connector acts as you, with your access, so do not share the key. If it is ever exposed, regenerate it on your profile and update the connector in your assistant.
Troubleshooting
| Problem | What to do |
|---|---|
| Every analytical question is refused | The property has no default analysis area. Set one in Lens Models. |
| The assistant does not see the TagPipes tools | Start a new chat after adding or reconnecting the connector. |
| A broad question goes to email or calendar | Use a Project or workspace with only this connector turned on. |
| The connector stopped working after you regenerated your token | Update the x-api-key header in your assistant with the new key. |