Orchestrate revenue agents with Workflows

Attio MCP: prompts and best practices

Work with your CRM in natural language using the Attio MCP.

Table of Contents

Available on all plans.

Available for all workspace members.

Attio MCP connects external AI assistants like Claude and ChatGPT to your Attio workspace, so you can search your CRM, update records, log notes, and manage tasks in natural language from wherever you already work with AI. This guide covers connecting Attio to your AI assistant, verifying the connection, and a library of role-based prompts for getting real work done, whether you run the whole company, RevOps, or your own book of business.

How the connection works

MCP (Model Context Protocol) is an open standard that lets AI assistants securely interact with other systems. Your AI assistant connects to Attio's MCP server, you sign in with your Attio account, and the assistant can then use a set of Attio tools: searching records, creating and updating records, working with notes and tasks, and reading emails and meeting data.

See the full list of the Attio MCP’s tools.

Three things are worth understanding before you connect:

  • It acts as you. The AI assistant authenticates as your Attio user through OAuth. It can only see and change what your own permissions allow.

  • You control what runs without asking. By default, your AI assistant asks for your approval before running most Attio tools. In your assistant's settings, you can allow specific tools, like read tools that search and view data, to run without asking. If you're in a managed workspace on your AI assistant, the available options may be restricted, e.g. write tools can’t be set to Always allow.

  • It works alongside Ask Attio, not instead of it. Ask Attio is AI inside Attio and needs no setup. MCP is for working outside Attio, where you can combine your CRM with other tools and data the assistant can reach. Learn more about when to use Attio MCP or Ask Attio.

Before you start

  • You need an Attio account on any plan. All workspace members can use the MCP.

  • You need an AI assistant that supports MCP connections, such as Claude or ChatGPT. Whether connectors are available can depend on your plan with that assistant, and if your AI assistant is managed by an admin, you may need them to enable the Attio connector.

  • Know which workspace you want to connect. If you belong to more than one Attio workspace, you choose one during sign-in.

Connect your AI assistant

Claude and ChatGPT connect through their built-in connector directories, and any other MCP-compatible assistant connects using the server URL https://mcp.attio.com/mcp. Follow the setup steps for your assistant in the Attio MCP developer documentation, signing in with your Attio account when prompted.

Once connected, Attio appears in your assistant's list of connected tools, and it can answer questions about your workspace.

Verify the connection

Confirm the assistant is talking to the right workspace with a read-only prompt in your AI assistant:

Which Attio workspace am I connected to, and what access level do I have?

Then try a real lookup:

Find the company record for [a domain you know is in your CRM, e.g. attio.com] and summarize what we know.

The assistant should name your intended workspace and return real record data. If it names a different workspace, disconnect and reconnect, then pick the right workspace during sign-in.

Prompt library

Copy any prompt and replace the values in [square brackets] with your own.

Some prompts pair Attio with another tool, like Slack, Linear, Intercom, or Apollo. To run those, your AI assistant needs a connection to that tool as well, and the tools named are examples, so swap in the ones your team uses.

Quick lookups to start with

Here are some single-line prompts that work as soon as you're connected.

Find the decision maker:

In Attio, who's the decision maker at [Meridian Labs]?

Find team conversations about a customer (with Slack):

What has the team said about [Brightline] in Slack over the last month?

If you manage your own pipeline

Reps with back-to-back calls use these prompts to keep their pipeline current without living in the CRM.

Prep for a call (with Intercom):

I have a renewal call with [Meridian Labs] at [2pm]. Pull the deal history, associated people, and open tasks from Attio, and check Intercom for their recent support conversations. Give me a one-page brief and flag anything that I should be aware of for the conversation.

Update after a call:

Review my last call with [Meridian Labs] in Attio. Move the deal to [Negotiation], log a note with the key points and objections, create follow-up tasks with deadlines, and draft the follow-up email for me to review.

Answer a product question mid-deal (with Linear):

[Meridian Labs] asked about [SSO support] on today's call. Check the status of that work in Linear and log a note on their Attio deal with what I can share and a realistic timeline.

One thing to know about email drafting: Attio MCP doesn't draft or send email. It supplies the context, like notes, transcripts, and deal history, and your AI assistant writes the draft in your conversation. From here you can:

  • Copy the draft: Paste it into your mail client, review, and send.

  • Connect your email to the assistant: If your assistant also connects to your mailbox, such as Claude with Gmail, it can place the draft there directly for you to review and send.

Either way, if your mailbox is synced with Attio, the sent email lands on the record automatically.

If you're a founder running sales solo

Use the MCP to keep the CRM current without living in it, and to grow your pipeline from what's already working.

Log a conversation from anywhere:

I met [Dana Okafor] from [Brightline] at an event today, their email is [[email protected]]. Create person and company records if we don't have them, note that [they're evaluating CRMs for a 12-person sales team], and set a task to follow up on [Thursday].

Find your next customers from your best ones (with Apollo): 

Find the [10] records with the highest [ARR] in Attio and take their associated companies. Use their domains to look them up in Apollo, then find [20] similar companies by industry, employee count, and location. Create company records in Attio for any we don't already have, with a task to research each.

If you run RevOps or admin the workspace

MCP lets you run data quality checks and ad hoc audits, and you can give your whole team leverage by standardizing the prompts that work.

Audit data quality: 

Show me open deals missing a value for Close date or Deal owner. List the attribute options for Deal stage first so we use the exact names.

Check support health before an outreach push (with Intercom): 

I'm about to run an expansion push on [companies with an open deal]. Check Intercom for any with unresolved or escalated conversations, I don't want to pitch anyone mid-complaint.

Build a targeted list: 

In Attio, build me a call list: companies in [fintech] with an open deal and no contact from us in the last [30 days]. Include the deal owner and the best person to reach at each.

Tip: The prompts that survive daily use are worth standardizing. Write the best ones into a shared doc, or save them as reusable prompts or skills in your AI assistant, so every rep runs the same reliable version instead of improvising.

Best practices for prompting

Here are a few prompting habits to help you get the best results from the Attio MCP:

  • Name records precisely. Use a full name plus company, or better, a unique value like an email address or domain.

  • Say which object you mean. "Create a deal for Brightline" is clearer than "add Brightline," which could mean a company, a person, or a list entry.

  • Use attribute names as they appear in Attio. If you're unsure, ask the assistant to list an object's attributes first, then reference the exact names and options.

  • Include a unique value when creating records. Provide an email address for person records or a domain for company records. Attio deduplicates automatically when a unique value is provided, so Attio can update existing records or create them if they don’t exist.

  • Ask for a change log. End write requests with "list everything you changed, with links to the records." Then spot-check the record in Attio while you're learning which phrasings work reliably.

Troubleshooting

  • The assistant says it can't see a record you know exists: Your Attio permissions apply to the connection too. Read access to objects and records can't be restricted per member, but list access can be. If a list is restricted for you, the assistant can't see its entries either.

  • Changes appear in the wrong workspace: The connection was authorized against a different workspace. Disconnect the Attio connector in your AI assistant, reconnect, and choose the correct workspace during sign-in.

  • Duplicate records after the assistant creates: Creation prompts without a unique value can produce duplicates. Include an email address for people or a domain for companies, and Attio matches to the existing record automatically.

  • Requests start failing after heavy use: MCP applies rate limits per workspace. Ask the assistant to slow down or run operations one at a time, then retry. Normal conversational use stays well within limits.

Related resources