Pipedrive MCP server: 5-minute setup for Claude and ChatGPT

Your CRM can now answer in plain English, but does it actually know anything worth asking about? Pipedrive shipped a native MCP server in June 2026, available to customers on every plan, and wiring it to Claude or ChatGPT takes about five minutes with no code and no developer. This guide gives the exact click path for both assistants, the prerequisites nobody mentions before you start, and the one question your assistant will answer confidently and wrongly.

What the Pipedrive MCP server does

The Model Context Protocol, documented at modelcontextprotocol.io, is an open standard that lets AI applications reach an external system through declared tools instead of a bespoke integration. Pipedrive hosts its own remote server, currently labelled Pipedrive MCP BETA in the connector setup, and presents it on its MCP feature page.

Because the server is built and maintained by Pipedrive, it queries live CRM data rather than a synced copy, and it does not depend on a community package staying updated. What that buys you day to day:

• Look up records without building a filter or opening the CRM.
• Create and update: move a deal stage, edit a contact, log an activity, convert a lead.
• Interrogate the pipeline: which deals have no activity scheduled, which are concentrated in one stage, which slipped twice.
• Chain small repetitive steps that would otherwise mean clicking through several records.

What you need before you connect

The server URL is the same in every case: https://mcp.pipedrive.ai/mcp. Interface labels move around, so Pipedrive's knowledge base stays the reference for the current wording.

How to connect Pipedrive MCP to Claude

Four steps, per Pipedrive's Claude setup guide:

1. Go to Customize, select Connectors, click the plus icon and choose Add custom connector.
2. Enter Pipedrive MCP BETA as the app name, paste https://mcp.pipedrive.ai/mcp in the Connection field, then click Add.
3. Sign in to Pipedrive if prompted, review the permissions listed in the Pipedrive pop-up, and click Allow and install.
4. Open a new chat and confirm the connector is enabled before you start.

One setting catches people out. Claude does not necessarily load every available MCP tool when a conversation starts, so if the assistant claims it cannot do something it clearly should, enable the option that loads all available tools in the connector configuration before assuming the server is at fault.

How to connect Pipedrive MCP to ChatGPT

Same server, one extra prerequisite, following Pipedrive's ChatGPT setup guide:

1. Turn on developer mode: Settings, then Apps, then Advanced settings, then enable Developer Mode.
2. Still in Settings, click Create app.
3. Enter Pipedrive MCP BETA as the app name and paste https://mcp.pipedrive.ai/mcp in the Connection field.
4. Complete the Pipedrive authorization, then open Tools or Apps depending on your ChatGPT version and check that Pipedrive MCP BETA is enabled and selected for the conversation.

Worth separating from the Pipedrive app for ChatGPT, which is a different product with its own plan requirements. The MCP route described here is the one that works from any compatible client.

The guardrails, and what they do not cover

Three things govern what the assistant can do. It inherits the permissions of the user who authorized the connection, so it sees exactly what that person sees and nothing more. Every action performed through an AI assistant is recorded in Pipedrive's change logs, so a bad write is traceable after the fact. And usage is metered: token limits depend on your plan, with the option to buy more.

What none of that covers is accuracy. A traceable wrong update is still a wrong update. Connect with a user whose visibility matches what you actually want the assistant to touch, then spend the first week asking questions you already know the answer to. That comparison tells you more than any permission matrix.

What it cannot tell you

An MCP server is a pipe, not a source. Ask your assistant which deals stalled on a pricing objection and it will query your deals faithfully, then answer from whatever the relevant field contains. In most Pipedrive accounts, that field is empty or holds a one-line note typed from memory two days after the call.

That is the ceiling. Stage, value and close date are structured and reliable. Why a deal is really stuck, who actually decides, what the competitor offered: those are said out loud in a conversation, and they reach the CRM only if someone put them there. An assistant reading half-filled records produces confident answers about incomplete data, and it will not warn you.

Filling the fields the assistant reads

Praiz addresses that at the source. It records and transcribes calls and meetings in over 100 languages, then configurable AI agents extract what was actually said and write it into the matching Pipedrive fields through the native integration: pain, budget, decision process, objections, competitor mentions, next steps. The rep sells, the fields fill themselves, and every MCP query afterwards has something real to read.

Praiz also exposes its own MCP server for the conversation layer itself. Run both in the same chat and you can cross pipeline state with call content: which deals in negotiation raised a pricing objection last week, and how the rep answered.

Across Praiz customer teams, internal data shows a CRM completion rate multiplied by 5 and +90% reliability on strategic CRM fields. Onboarding and field mapping are done with you rather than left to your admin, which is usually what decides whether any of this survives the first month.

Running another CRM?

Pipedrive is one of three CRMs with a native server today. For the wider picture, including HubSpot and Salesforce and the options for CRMs with no native connector, see our guide to the CRM MCP server landscape, or the detailed walkthrough of the HubSpot MCP server.

Frequently asked questions