Attio + Claude: run your pipeline from a chat window

Last updated 17 August 2026 · Nacho Lafuente, Attio Certified Partner

Your CRM already knows which deals went quiet this month. Getting that out of it takes four filters and a saved view you will never open again. The connection itself takes about two minutes. The part that decides whether this is a party trick or a daily tool is entirely in the questions you ask afterwards, so most of this page is those questions, written out so you can paste them.

How to Connect Attio to Claude (2 Ways, 90 Seconds)

Direction

Claude ↔ Attio (read and write, via MCP)

Stack

Attio MCP, Claude Desktop, Claude web

The what

What the connection actually gives you

Attio ships an MCP server. Claude connects to it and gets scoped access to your workspace: it can search and read records, notes, tasks and list entries, and it can create and update records when you grant write access. What it cannot do is drive the Attio UI. It will not build a view, configure an automation, reorder a pipeline or change your schema. Those stay manual. Anyone selling you "Claude configures your CRM" is describing something the MCP does not expose. So the honest framing is: conversational read and write over records you already have, not an admin who sets the workspace up for you.

The how

How to connect it

1
Open Claude, go to Settings, then Connectors.
2
Add Attio. Claude redirects you to Attio to authorize the connection.
3
Read the access screen rather than clicking through it. Whatever you approve here is what Claude has in every conversation, forever, not just the one you are about to have.
4
Start read-only. You can add write access later, once you have watched it be right about your data for a couple of weeks.
5
Prove it works on something trivial ("how many companies are in the workspace") before you trust it with anything real.
6
Alternative path: from inside Attio, go to Apps and Integrations, search Claude, and connect from that side. Same result, same authorization screen.

Under the hood

The five prompts worth saving

Pipeline hygiene: "Which deals have had no activity in the last 21 days? Group them by owner." Every sales lead wants this weekly and nobody builds the view, because by the time you build it you have already answered it once.
Context recovery: "Read the last three notes on this company and tell me what we promised them." Six weeks after a call nobody remembers what was committed. The answer is sitting in the CRM in a form no human will re-read.
Revenue gaps: "Find every company in the pipeline with no decision-maker contact attached." Not a reporting question. That is a list of deals about to stall, handed to you before they stall.
Loss analysis: "Summarize every deal we lost last quarter and cluster the stated reasons." People buy a BI tool for this. You already have the data.
Execution: "Draft a follow-up to this person using the last call notes, in my voice." The only one on the list that writes rather than tells, and the one that saves the most time per day.

Hard-earned lessons

What nobody mentions in the demo

It does not know what it does not know. If half your deals have no owner and your stages mean different things to different reps, you get a fluent, confident, completely wrong answer that looks exactly like a right one. The AI layer is only as good as the schema underneath it.
Write access is real access. A confidently wrong bulk update is a bad afternoon. Read-only first, always.
Scope is granted once, not per question. Every conversation carries the access you approved at setup, not the access you meant to grant for that one query.
No UI control. Views, automations, Kanban status attributes and formulas are not exposed over MCP. Expect to build those by hand.
Large result sets get truncated before the model sees them. For anything counting across thousands of records, verify against a real Attio report rather than trusting the number in the chat.

Case study

Nordic PE fund, 12 seats

Problem

The team ran on Claude all day and their CRM not at all. Weekly pipeline review meant one person exporting to a spreadsheet every Monday morning.

Solution

Attio MCP connected read-only, plus five shared prompts written once and circulated instead of everyone inventing their own phrasing.

Outcome

The Monday export stopped. The stalled-deal question now gets asked mid-week by whoever wants it, not once a week by whoever drew the short straw.

FAQ

Questions we get

No. The Attio MCP exposes records, notes, tasks and lists. Views, automations, Kanban status attributes and formula attributes are UI-only, so they stay a manual build. This is the single most common misunderstanding about the integration.

Eventually, yes, but not on day one. Run read-only until you have seen it answer correctly about your own data for a couple of weeks. Write access means a misread prompt can update records in bulk, and undoing that in Attio is manual.

Whichever your team already has open. Claude tends to be better at chaining several CRM operations in one request. ChatGPT wins on the thing that matters more in practice, which is that everyone already has it. The full comparison is on the ChatGPT page.

Connectors require a paid Claude plan. The Attio side needs a seat with access to the objects you want to query; expert access grants work too, so a consultant can connect without consuming a paid Attio seat.

Data passed into a conversation is processed under Anthropic's commercial terms and is not used for training on paid plans. For EU funds with strict policies, the practical control is scoping: connect a workspace or grant access limited to the objects you are willing to have summarized in a chat window.