Set up ImpactPilot MCP

This guide explains how to connect ImpactPilot to an AI assistant — Claude, ChatGPT, Gemini, or Copilot — so you can ask questions about your customer signals in plain English.

The connector uses an open standard called MCP (Model Context Protocol). Support for custom MCP connectors, and the menus to set them up, differ a lot between assistants and change often, so treat the exact wording below as a snapshot rather than gospel — if a menu has moved, your assistant's own help center is the tie-breaker.

Before you start

You need:

  • An ImpactPilot account that you can log into normally.
  • An AI assistant that supports custom MCP connectors, and a plan/account type that unlocks it (see each section below — this is usually a paid plan, and sometimes needs an admin).
  • The connector URL: https://mcp.impactpilot.io/mcp  

Set it up in Claude

Requirements: the Claude desktop app or claude.ai in your browser. A paid Claude plan is generally required for custom connectors.

  1. Open Claude — either the desktop app or claude.ai in your browser.
  2. Click your name or profile picture in the bottom-left corner, then choose Settings.
  3. Open the Connectors section.
  4. Click Add custom connector (you may need to scroll to the bottom of the list of existing connectors). Fill in the two fields:
    • Name: ImpactPilot  
    • URL: https://mcp.impactpilot.io/mcp  
  5. Click Add.

Signing in

  1. Claude will now open an ImpactPilot login window in your browser. Log in the way you normally log into ImpactPilot (through HubSpot). If you're already logged in, you'll skip straight past this.
  2. ImpactPilot will ask whether you want to give Claude access to your data. Click Approve (or Allow).
  3. The window closes and returns you to Claude. The ImpactPilot connector should now show as Connected.

That's it — you don't have to repeat this. Claude remembers the connection.

Check that it worked

Start a new chat and type:

Using ImpactPilot, who am I?

Claude should reply with your name, your email and your organisation. The first time it uses the connector it may ask for your permission to do so — allow it. You can tick the option to allow it for the rest of the conversation so you're not asked again for every question.

Set it up in ChatGPT

Requirements: custom MCP connectors sit behind Developer mode, which needs a Plus, Pro, Business, Enterprise, or Edu plan. The free plan can't add custom connectors. On a Business, Enterprise, or Edu workspace, a workspace admin may need to turn Developer mode on before you can use it (Workspace Settings → Permissions & Roles → Connected Data → Create custom MCP connectors) — if you don't see the toggle at all in step 1, that's usually why.

  1. In ChatGPT on the web, go to Settings, find the Connectors (or Apps) section, open Advanced settings, and switch on Developer mode.
  2. Go to Settings → Connectors and click Create (or Add custom connector).
  3. Fill in the fields:
    • Name: ImpactPilot  
    • Description: something like "Reads customer signals, accounts and impact drivers from ImpactPilot" — ChatGPT uses this to decide when to call the connector, so keep it specific.
    • MCP server URL: https://mcp.impactpilot.io/mcp  
    • Authentication: choose OAuth.
  4. ChatGPT will open the ImpactPilot login window. Log in through HubSpot as usual, then click Approve when ImpactPilot asks for access.
  5. Once it shows as connected, start a new chat and turn the ImpactPilot connector on from the tools/connector icon near the message box — with Developer mode, connectors aren't always on by default in every chat.

Check that it worked

Ask:

Using ImpactPilot, who am I?

The first time ChatGPT calls a tool from the connector, it will usually ask you to confirm the call before it runs — approve it.

Set it up in Gemini (Admins only)

Requirements: this is for teams on Gemini Enterprise (formerly Duet AI / Gemini for Google Workspace, business/enterprise tier) — not the consumer Gemini app. You'll need admin access to your team's Connected Apps settings, or an admin who can complete this for you. Unlike Spark, there's no personal-account, US-only, or 18+ requirement — this is managed at the organization level.

  1. Sign in to Gemini Enterprise at business.gemini.google.com.
  2. Click Settings & help.
  3. Select your Team, then go to Manage team → Connected apps.
  4. On the Connected apps page, click Add MCP Server.
  5. In the MCP info section, enter the server details:
    • Name: ImpactPilot
    • MCP server URL: https://mcp.impactpilot.io/mcp
  6. In the Authentication settings section, fill in the required fields and follow the on-screen steps to log in through HubSpot and approve access, the same as with Claude.
  7. Click Add. The server will appear on the Connectors page — note that it's disabled by default, so click to enable it and make it visible to your team.

Set it up in Copilot

Unlike Claude, ChatGPT, and Gemini, Microsoft Copilot doesn't currently have a "paste a URL into your own settings" option for an individual end user. Connecting a custom MCP server to Microsoft 365 Copilot goes through Copilot Studio (Microsoft's agent-building tool) or the Microsoft 365 admin center, and normally needs a maker or IT admin to do the setup and then make it available to you and your colleagues. If that's not something you have access to, this is one to hand to your admin rather than do yourself.

For whoever does the setup:

  1. In Copilot Studio, open (or create) the agent that should use ImpactPilot, and go to its Tools page.
  2. Select Add a tool → Model Context Protocol → New custom connector (or use the MCP onboarding wizard if it's available on your tenant).
  3. Enter the connector details:
    • Name: ImpactPilot  
    • MCP server URL: https://mcp.impactpilot.io/mcp  
    • Transport: Streamable HTTP
    • Authentication: OAuth 2.0
  4. Save, then authorize the connection — this opens the same ImpactPilot login (through HubSpot) and approval screen described above.
  5. Add the tool to the agent, and publish or roll it out to the users or group who should have access (for example, to Microsoft Teams or the Microsoft 365 Copilot app).

Once it's rolled out, users can ask their Copilot agent things like "Using ImpactPilot, who am I?" the same way as in the other assistants.

Good to know

These apply no matter which assistant you use:

  • It only reads. Nothing the assistant does can change your signals or any other data in ImpactPilot.
  • Long lists get shortened. If you ask a very broad question, the assistant may only read the first 25 signals in detail. For big questions, prefer counting questions ("how many per customer") over "list everything".
  • Always sanity-check numbers you'll report onward. The assistant reads real data, but it writes the summary itself. Use the signal links to verify anything important before sending it to a customer or your leadership.
  • Tool confirmations are normal. Most assistants will ask you to approve the first tool call in a conversation (and sometimes every call). This is the assistant's own safety behaviour, not an ImpactPilot error.

If something goes wrong

"My assistant doesn't seem to know about ImpactPilot." Check that ImpactPilot shows as connected in your assistant's connector settings. Also make sure the connector is switched on for the specific chat you're in — in most assistants there's a tools or connector icon near the message box where you can turn it on, separate from having it connected account-wide.

"It says it isn't authenticated / asks me to log in again." Your login session expired. Go back to the connector settings, find ImpactPilot, and reconnect (or remove it and add it again). You'll go through the same approve screen as during setup.

"The login window opened but nothing happened." Make sure your browser isn't blocking pop-ups for your assistant, then try connecting again.

"It can't find a customer I know exists." Try the exact name as it appears in ImpactPilot. Partial names usually work, but nicknames and abbreviations may not. If you know the customer's HubSpot record ID, you can give that instead.

"It says there are no signals but I can see some in the app." By default it only looks at open (unresolved) signals from your own accounts. Ask again with "including resolved ones", or "across the whole organisation" if you have permission to see that.

"I don't see a custom connector / developer mode option at all." This usually means your plan or account type doesn't unlock it (see the requirements at the top of each assistant's section above), or — for ChatGPT Business/Enterprise, Gemini Enterprise, and Copilot — that an admin needs to enable or set it up first.

Still stuck?

Contact your ImpactPilot support, and mention which assistant you're using, which step of this guide failed, and what the error message said.