Sistava

How to Connect Pipedrive to an AI Assistant With MCP

How-to — — by Mahmoud Zalt

Connect Pipedrive to Claude or ChatGPT with its native MCP server. Setup steps, what the assistant can create and update, permissions and rate limits.

Pipedrive is the CRM people pick because it stays out of the way, so the appeal here is obvious. Ask what moved this week, get an answer, skip the report builder. That part works and takes about three minutes to set up.

The half that deserves more thought is that the same connection can create and update records. A beta integration with write access to the system your revenue depends on is not a reason to avoid it, but it is a reason to be deliberate about the order you enable things in.

What is the Pipedrive MCP server?

It is Pipedrive's own connection point for AI assistants, built on the Model Context Protocol, an open standard that lets an assistant reach outside tools and data. You connect through an OAuth sign-in, with no middleware and no developer work. Pipedrive describes it as working with any AI assistant that supports custom MCP connections, and publishes setup guides for both Claude and ChatGPT.

On capability, Pipedrive lists searching deals, contacts, organisations and leads, retrieving pipeline details, and analysing sales activity on the reading side. On the writing side it lists creating and updating deals, contacts and activities, converting leads, and scheduling follow-ups.

The security model is the part worth reading twice. Pipedrive states that MCP does not automatically grant access to your data, that access is managed through authentication, permissions and authorisation settings, and that every action follows the user's existing permissions and is recorded in Pipedrive's change logs. Your assistant is exactly as powerful as the account that signed in.

How do you connect it to Claude?

Through the custom connector screen, using the MCP server URL. You need permission to install and use connectors in Claude, and permission to install apps in Pipedrive. The whole thing takes a couple of minutes and needs no configuration files.

Adding Pipedrive to Claude

  1. Open the connector settings — In Claude, go to Customize, then Connectors, click the plus button and choose Add custom connector.
  2. Name it and paste the URL — Pipedrive's guide uses Pipedrive MCP BETA as the app name, with the server URL https://mcp.pipedrive.ai/mcp.
  3. Sign in to Pipedrive — Authenticate when prompted, then click Allow and install on the permissions confirmation.
  4. Load the full toolset if you need it — Pipedrive notes that Claude may not load every available MCP tool at the start of a conversation, and that you can enable the setting that loads all available tools in the connector configuration.
  5. Ask before you edit — Spend the first sessions on questions you can verify by hand. If it misreads a pipeline or a custom field now, you want to know before it starts writing.

For ChatGPT the path differs slightly. Pipedrive's guide has you enable developer mode in ChatGPT's Advanced settings, click Create app, use Pipedrive MCP BETA as the name, and enter the same connection URL before authorising in Pipedrive. Same server, same permission model, different door.

What are the rate limits you will actually hit?

Pipedrive meters API usage with a daily token budget rather than a simple request count. The formula it publishes is 30,000 base tokens multiplied by your plan multiplier and your number of seats, plus any purchased top-ups. The multipliers are one for Lite, two for Growth, five for Premium and seven for Ultimate, and the budget resets at the server's midnight, which may not match yours.

WhatHow it works
Daily budget30,000 base tokens times plan multiplier times seats, plus top-ups
Plan multipliersLite 1, Growth 2, Premium 5, Ultimate 7
Burst, OAuth apps80, 160, 400 and 480 requests per user per 2 seconds by plan
Search endpoints10 requests per 2 seconds on every plan
Sample costsFetch one record 2 tokens, list 20, update 10, search 40
When exhaustedRequests are rejected with a 429 until the next reset

For a person asking questions, this is a non-issue. It matters when someone decides to have an assistant sweep every deal in the pipeline, because searches are the most expensive calls in the table and a broad sweep can spend a small team's whole daily budget before lunch.

Why access is the easy problem

Every step above solves the same thing: getting an assistant to your CRM. None of it solves what breaks pipelines, which is that the records were already incomplete, duplicated or labelled by three people who each meant something different by qualified.

Point an assistant at that and you get answers that read beautifully and mislead quietly. Give it write access and the mess grows faster than anyone can review it. Connecting takes three minutes. Deciding what good data looks like is the work that actually pays.

That is the difference between a connector and a colleague, and it is why Sistava works the way it does. You hire an AI Employee for a defined job with a named field list, so it keeps the record current as part of doing the work rather than waiting for someone to open a chat window and ask.

Whichever way you go, keep the same order. Read-only for a week, then one narrow write job, then widen slowly. In a beta integration that order matters even more, because the behaviour you tested last month is not guaranteed to be the behaviour you get next month.

What this setup will not do

It will not run without you. An assistant with a connector responds to prompts, so nothing happens on the days nobody asks. If the goal was to stop remembering, a connector is not the tool for that job.

It will not standardise your data. Pipedrive's permission model controls who can reach what, not whether the values make sense. Two reps using different stage definitions will still produce a pipeline that cannot be trusted, whether a human or an assistant typed it.

And it will not stay still. This is labelled beta, so expect screens, tools and behaviour to change without much warning, and check the change log after anything surprising. If you want to compare that with something that runs the job on its own, Sistava is free to look around before you decide.

A practical closing suggestion. Connect it with your own account rather than a shared admin login, because the permission model is the guardrail and a shared login with full rights removes it entirely. Then spend a week asking questions you already know the answer to. That week is the cheapest quality check you will ever run on your own CRM.

Frequently asked questions

FAQ

What is the Pipedrive MCP server URL?

Pipedrive's knowledge base gives https://mcp.pipedrive.ai/mcp as the connection URL, used both when adding a custom connector in Claude and when creating an app in ChatGPT's developer mode. Because it is in beta, check the current guide before saving it anywhere permanent.

Can the AI create and update Pipedrive records?

Yes. Pipedrive lists creating and updating deals, contacts and activities, converting leads and scheduling follow-ups among the supported actions, alongside the read side. Everything runs under the permissions of the account that signed in, so those permissions are your real control.

Do I need admin rights to set this up?

You need permission to install apps in Pipedrive, and on the Claude side permission to install and use connectors. Whether that means an admin depends on how your Pipedrive account is configured, so check your own role before you start.

Can I see what the AI changed?

Yes. Pipedrive states that actions taken through an AI assistant are recorded in its change logs, which is exactly what you want before granting write access. Check the log during the first weeks rather than waiting for something to look wrong.

Will this use up my API limits?

It draws on the same daily token budget as any other integration, calculated as 30,000 base tokens times your plan multiplier times your seats. Ordinary question and answer use is nowhere near the ceiling, but broad searches are the most expensive calls and a large sweep can spend a lot of it quickly.

Is it safe to use a beta integration with my CRM?

It is reasonable if you stay on the reading side while it settles. The risk with a beta is not usually a dramatic failure, it is quiet changes in behaviour between releases. Keep write permissions narrow, watch the change log, and widen only once things have been stable for a while.

Sources