Passer au contenu
TL;DR: What Is Pipedrive MCP?
Pipedrive MCP is the official Model Context Protocol server from Pipedrive, launched in June 2026. It lets AI assistants like Claude and ChatGPT read and update your CRM using natural language. After testing it personally, I can confirm it’s actually pretty easy to use too.
It’s available on every Pipedrive plan, connects over OAuth with no code required, and only touches data you already have permission to see. Token limits apply, and make sure you enable checks for write access.
Read on for more details!
Pipedrive MCP is Pipedrive’s official Model Context Protocol server: a secure bridge that lets AI assistants like Claude and ChatGPT query and update your CRM in natural language. Pipedrive launched this official version on June 30, 2026, but be warned, plenty of community versions also exist and it’s quite easy to connect to the wrong one.
Pipedrive’s MCP is built so that you don’t need to do the legwork of searching through CRMs yourself. Ask your AI assistant to do it and it’ll cover the whole database in a fraction of the time on your behalf. Whether you’re looking for a specific deal, or a general overview, Pipedrive MCP is a useful tool for automating admin tasks.
Here’s what Pipedrive MCP actually does, how to connect it, what it costs, and the security questions you should consider before you give a chatbot free access to your pipeline.
What Is Pipedrive MCP?
Pipedrive MCP is a connector that exposes your Pipedrive CRM to AI assistants through the Model Context Protocol. If that’s a load of gibberish to you, it simply means that instead of clicking through the Pipedrive interface, you type (or say) what you want, and a compatible AI client fetches or changes the data for you. Depending on the restrictions you set, the AI can work with deals, contacts, activities, notes, and more.
What Is the Model Context Protocol (MCP)?
MCP is an open standard, introduced by Anthropic in November 2024, that gives AI models a consistent way to connect to outside tools and data. The usual analogy is a USB-C port for AI: one standard plug instead of a custom cable for every app.
Before MCP, wiring an AI assistant into a tool like Pipedrive meant bespoke code. MCP replaces that with a shared protocol, so any MCP-compatible client, like Claude, ChatGPT, or Cursor, can talk to any MCP server without a one-off integration.
I covered the full picture in my Model Context Protocol guide. Here’s the short version: MCP is swiftly becoming the default way AI interacts directly with the software you already use.
Is Pipedrive MCP Official or Community-Built?
There are both official and community-built MCP servers for Pipedrive. The official one is the one that matters. Pipedrive’s official MCP server connects over OAuth, respects your existing Pipedrive permissions, and logs every action in the change log.
Search “Pipedrive MCP,” though, and you’ll also hit a pile of third-party and open-source servers on GitHub and npm, some with far bigger tool lists than the official one. Those are independent projects, not Pipedrive products, and they carry their own setup and security trade-offs. I’ll cover the main differences below, but the general rule of thumb is that unless you have a specific reason to self-host, the native server is the safer option.
How Does Pipedrive MCP Work?
When you get down to the nuts and bolts of it, Pipedrive MCP follows the standard MCP client–server model. Your AI assistant is the client; Pipedrive is the server. In simple terms, it means Pipedrive allows the AI assistant to act within the app as if it were a user, giving it access to “tools” like searching for a deal or creating a contact. This all hinges on the access restrictions you set in advance.
Start by connecting Pipedrive to your AI client once, through a secure OAuth sign-in. There’s no API keys required and no complex dev stuff. From then on, when you ask “What are my open deals with Acme Corp?”, the assistant picks the right tool, sends the request to Pipedrive’s server, and gets live data back. Ask it to move that deal to Proposal and it calls a write tool instead.
It’s worth remembering that the assistant only sees and edits data your Pipedrive user already has permission to touch (MCP inherits your access). Additionally, every action lands in Pipedrive’s change log for auditing. It reads and writes.
What Can You Do With Pipedrive MCP?
Quite a lot. Almost everything you can do manually, you can now also do via natural language. Essentially, if you have a task to do in Pipedrive, you can simply ask your AI assistant to do it and you’ll have the result in under a minute, freeing up your time to focus on more important tasks.
Read and Write Actions
On the read side, you can search and retrieve deals, contacts (persons), organizations, leads, activities, and notes, and pull pipeline and stage details. Handy prompts:
- Find every open deal that hasn’t moved in two weeks.
- Surface all overdue activities.
- Retrieve everything the CRM knows about [contact] before our call at 12.
On the write side, you can create and update deals, contacts, organizations, and activities, add and edit notes, convert leads into deals, and schedule follow-ups. So “create a contact for Jane at Acme, link her to the account, and book a call for Monday” becomes one instruction instead of five screens.
It’s most useful to combine read and write actions: read the state, decide, then write the change. It can all happen in one conversation. Just make sure you have it set so you can verify each read and write before confirming. See the screenshot below for a glimpse of how this looks in action.
The Pipedrive MCP Tools (Naming Convention)
Pipedrive’s MCP tools follow a predictable naming pattern. They’re grouped by CRM object: activities, deals, persons, organizations, leads, pipeline stages, and notes. They’re also prefixed by action: getX retrieves a record or list, addX creates a new record, updateX changes an existing one, and searchX finds records matching a query.
So getDeals lists deals, addPerson creates a contact, updateOrganization edits an account, and searchLeads hunts down leads. You rarely need to name these yourself as the assistant picks the tool on your behalf, but knowing the pattern helps when you’re working out why it did (or didn’t) do something.
Example Prompts
A few prompts that work well once you’re connected:
- Search: “Show me all open deals that haven’t been updated in two weeks.”
- Retrieve: “Show me everything we know about the contact at Acme Corp.”
- Update: “Move the Acme deal to Proposal and set its value to $12,000.”
- Schedule: “Create a follow-up call with Jane for next Monday at 10 a.m.”
- Meeting-to-CRM: “Here are my notes from today’s call — update the relevant deal and add the summary.”
That last one can be streamlined even further. More on that below.
How to Set Up Pipedrive MCP
Setup was surprisingly quick. It only took me a few minutes before I was able to use it. All you need is a Pipedrive account and an AI client that supports custom MCP connections. The connection itself is a one-time OAuth sign-in. After that, the tools show up in your assistant like this.
How to Connect Pipedrive MCP to Claude
Using Claude as an example, the easiest route is to open Claude’s Connectors page, hit Browse connectors, and search “Pipedrive MCP”. The official connector is already listed, so you don’t need to build a custom one. Pipedrive’s own guide still walks you through adding a custom connector, which is fine, but unnecessary. If you try to add it manually after connecting, Claude just tells you it’s already there.
Click Connect and Claude opens a tab asking you to sign in to your Pipedrive account, then accept the permissions. Approve them and Bob’s your uncle. You can start asking questions straight away. Just double check that Pipedrive is toggled on in the connectors menu under the message box (as seen in the screenshot above). The whole takes just a few minutes, with not a single line of code needed.
If your AI client doesn’t list Pipedrive for some reason, you can add a custom connector. You’ll have two options:
- Name: Pipedrive MCP
- Official server URL:
https://mcp.pipedrive.ai/mcp.
The screenshot below shows the “Add custom connector” screen.
tl;dv’s Claude MCP guide covers the general connector flow, and Dani’s Claude dashboard experiment shows what a connected pipeline looks like in practice.
How to Connect Pipedrive MCP to ChatGPT
The ChatGPT route is similar but lives in developer settings. You’ll typically need a plan that supports custom connectors, then you create a Pipedrive MCP connector, enter the connection details, and authorize Pipedrive access.
Once enabled in a conversation, ChatGPT can search and edit your CRM the same way Claude can. Because access follows your existing Pipedrive permissions, connecting from ChatGPT doesn’t hand it anything your own login can’t already do. If a tool doesn’t appear, the usual culprits are a missing connector permission or an authorization that didn’t complete. Try to reconnect and re-authorize.
Which AI Assistants Support Pipedrive MCP?
Officially, Pipedrive MCP works with any AI assistant that supports custom MCP connections. It has dedicated setup guides for Claude and ChatGPT, while other clients like Cursor can connect via the server URL. Capabilities vary slightly by client, since they each have different ways of utilizing tools through MCP, but the underlying server is the same.
Pipedrive MCP Pricing: Is It Free?
Pipedrive MCP is included on all plans, so you’ll already have it within your Pipedrive subscription. There’s no separate “MCP plan”. You are, however, metered on usage: each plan comes with a token allowance for MCP. You can buy additional tokens if you run out, but how much they cost is conveniently absent from Pipedrive’s website copy.
So “is it free?” sort of depends. It’s included at no extra cost, but it’s not unlimited. The base cost is whatever you already pay for Pipedrive. The variable cost is tokens, which scale with how much you actually use the MCP.
One cost people often forget is the AI client itself. Custom MCP connectors in ChatGPT generally require a paid plan on that side, so budget for the assistant as well as the CRM. I used it with Claude on a free account, but you obviously get limited tokens when you’re not on a paid subscription, so you’re more likely to run out when using connectors.
If usage-based AI billing is new to you, you might find my Claude enterprise pricing breakdown useful.
| Cost Component | What It Is | Included or Extra |
|---|---|---|
| Pipedrive subscription | Required — any Pipedrive plan works | Whatever you already pay |
| Pipedrive MCP access | Available on every plan; there's no separate MCP plan to buy | Inclus |
| MCP token usage | Each plan comes with a token allowance; buy more if you run out | Limits vary by plan |
| AI assistant (Claude / ChatGPT) | Custom MCP connectors generally require a paid plan on the AI side | Separate cost |
Is Pipedrive MCP Secure?
Mostly, yes, but as always, it’s more complicated than that. On the reassuring side, Pipedrive MCP authenticates over OAuth, inherits your existing user permissions (the AI can’t see records you can’t), and records every action in the change log for auditing. Pipedrive says the connection follows its own security standards, and your AI provider’s privacy policies apply on top.
When it comes to write access though, you’ll still want to be careful. Letting an AI assistant change CRM data can be risky if it’s left to its own devices. Simon Willison calls it the “lethal trifecta”: access to private data, exposure to untrusted content, and the ability to communicate externally. An AI with all three can be tricked by a malicious instruction hidden in, say, a lead’s message or a meeting note. This is called prompt injection. The AI cannot discern whether the instruction came from you or not, it just follows orders. CRM connectors have already featured in real prompt-injection demonstrations so this isn’t a complete fantasy.
But that’s not a reason to avoid Pipedrive MCP. It just requires the right roll-out. The client actually does a lot of the leg work for you here: in Claude, for example, the connector splits Pipedrive’s tools into 29 read-only and a similar amount of write groups. This lets you set each one to run automatically, ask first, or stay blocked.
It’s usually best to start with read-only and enable writes once you trust the workflow. Make sure you require confirmation before destructive or high-value changes and connect only the AI clients you actually vet. Use the change log to review what the assistant did afterwards. Admins should treat AI access like any other integration: scoped, monitored, revocable.
Dani’s overview of the best MCP servers dives into what makes a good one.
Native Pipedrive MCP vs Third-Party MCP Servers
Because MCP is an open standard, anyone can build a Pipedrive server, and plenty have. Search results are full of open-source projects. Some of these expose 75, 100, even 150-plus tools, well beyond the official server’s scope. So which should you use?
The official server is the right call for most people. It’s built and maintained by Pipedrive, connects over OAuth with no code, inherits your permissions, and logs to the change log. The trade-off is that you get less tools and token limits to consider.
Third-party servers tend to authenticate with a Pipedrive API token rather than OAuth, run self-hosted (you install and maintain them), and often cover more of the API surface. That can be useful for developers building custom agentic workflows, but you are responsible for the security of it. An over-permissioned server reachable over the internet is a real risk, and community projects vary wildly in terms of maintenance and vetting.
At the end of the day, unless you’re an expert, the official Pipedrive MCP is going to be the simplest, safest, and most reliable option. Use third-party only when you have a specific need that the official server can’t meet, and only then when you know what you’re doing.
| Fonctionnalité | Official Pipedrive MCP | Third-Party / Open-Source |
|---|---|---|
| Built & maintained by | Pipedrive | Independent developers |
| Authentication | OAuth sign-in | Pipedrive API token |
| Hébergement | Hosted by Pipedrive | Self-hosted (you run it) |
| Configuration | No code, connect in minutes | Manual install and config |
| Tool coverage | Curated core objects | Often broader (75–150+ tools) |
| Permissions | Inherits your CRM access | You configure it yourself |
| Audit trail | Oui | Variable |
| Meilleur pour | Most teams | Developers with a specific need |
Pipedrive MCP vs the Pipedrive API (and Zapier)
Pipedrive MCP doesn’t replace the Pipedrive API or your existing automations. They’re two different things and they sit alongside each other solving different problems.
The API is for deterministic, high-volume, coded integrations: sync ten thousand records nightly, build a custom app, wire up a data pipeline. It does exactly what you program, every time. MCP is for conversational, AI-driven work: ad-hoc questions, one-off updates, judgment calls phrased in natural language. It’s flexible because it isn’t rigidly scripted. This also means it’s less suited to bulky jobs.
Zapier is more of a middleman. It’s great for fixed “when X, then Y” rules, but you still define the rule up front. MCP, on the other hand, doesn’t require a pre-built workflow. You just ask what you want in words that make sense to you.
In practice, teams use all three: the API and Zapier set up big, hard rules that you need to execute on a regular basis, MCP lets you read and write remotely as a one-off operation here and there.
If all you want is call data flowing into Pipedrive automatically, the native Pipedrive integration with tl;dv may be all you need.
Using Pipedrive MCP With an AI Meeting Assistant
This is where some people get confused. Pipedrive MCP gives an AI assistant access to your CRM, but it doesn’t generate the meeting intelligence worth putting there. These are two different jobs that give more together.
Pipedrive MCP is reactive. It waits for you to open Claude or ChatGPT and ask. An AI meeting assistant like tl;dv is proactive: it joins your calls, records and transcribes in 40+ languages, and produces structured notes and action items automatically, even if no one opens a chat afterwards. tl;dv even runs its own MCP server, so meeting data across any platform (Zoom, Google Meet, Teams, and more) is queryable the same way CRM data is.
Chain them together and your afternoons start to become a lot more free. The call happens, tl;dv captures what was said and decided, and its native Pipedrive CRM meeting integration automatically fills the exact CRM fields you want it to with data from the call. tl;dv’s guide to automate CRM logging from sales calls walks through the hands-off version.
The best way to stack these is to use Pipedrive MCP to talk to your CRM, and the tl;dv integration with Pipedrive to automatically fill your CRM. That way, you can finish your call, open up Claude and query your CRM and the call data will already be in place.
Limitations and Troubleshooting
While Pipedrive MCP is mostly straightforward, there are a few caveats to consider. Token limits mean heavy users can run out fast, and without clear pricing for additional tokens, it makes it difficult to plan. The official server’s tool coverage is limited, so niche objects or bulk operations might not be available. Developers who need these tools can use the API or a third-party server in these cases.
Another thing to keep in mind is that MCP is conversational. Results are always going to depend on the clarity of your prompts. Vague requests get vague actions.
There are some common fixes to apply: if a tool doesn’t appear, confirm the connector is authorized and all tools are enabled. If the AI can see some records but not others, that shows your permissions are working. If an action fails, check you’re on a plan and role with the right access, then reconnect and re-authorize.
For the meeting side of MCP, my guides on Zoom MCP and Google Meet MCP cover the same stuff but from the other tools’ perspectives.
Connect to Pipedrive MCP
Pipedrive MCP is an official upgrade to how you interact with your CRM. You can now ask for what you want in plain language and let an AI assistant fetch or change it. It’s on every plan, quick to connect, and scoped to your own permissions.
The parts you need to think about are also the parts that Pipedrive is most vague about: metered tokens and how much they cost. You’ll also want to consider the risk of letting AI write to a live CRM. It’s always best to start with read-only permissions and expand as you familiarize yourself with it.
FAQs About Pipedrive MCP
Is Pipedrive MCP free?
It’s included on every Pipedrive plan at no extra base cost, but usage is metered by token limits, and you can buy more tokens if needed. You’ll also need a paid AI client for custom connectors.
Do I need a paid Claude or ChatGPT plan to use Pipedrive MCP?
Generally, you’re better off with one, yes. Custom MCP connectors in Claude and ChatGPT typically require a paid plan on the AI side to access the full suite of features. That cost is separate from your Pipedrive subscription, so budget for both.
Not to mention, the free plan will eat up credits very quickly when using connectors.
Does Pipedrive MCP work with Salesforce or HubSpot MCP too?
They’re separate servers, but a capable AI client can connect to more than one at once — so an assistant could read from Pipedrive and another CRM in the same conversation, each still bound by its own permissions.
Is Pipedrive MCP still in beta?
Pipedrive launched its native MCP server as a live feature available on all plans in June 2026, not a closed beta. Pipedrive has signaled it plans to expand MCP with more tools and workflows over time.
Does Pipedrive MCP support custom fields?
Custom field support varies by tool and object. The official server focuses on core objects, so if you rely heavily on custom fields or niche data, verify coverage for your setup or use the API for those specific jobs.
Can Pipedrive MCP delete records?
Destructive actions like deletion are the ones to gate. Best practice is to keep the AI read-and-create only, require confirmation for high-stakes changes, and use the change log to review anything the assistant does.
