Attio MCP: How to Connect Claude and AI Agents to Your CRM (2026)
•
Attio MCP is Attio's official, hosted MCP server. It connects your Attio CRM to Claude, ChatGPT, Cursor, Notion, and any other MCP client through a secure OAuth login, so AI agents can search, read, and update your records through plain conversation. It shipped in February 2026. This guide shows you how to connect it and where it earns its keep.
The bigger shift matters more than the setup. Your CRM stops being a place you visit and becomes a surface your AI agents operate. Research a market in Claude, then have it create the companies, add them to a list, and draft the outreach, without a single tab switch. That is what AI-native GTM looks like in practice, and it is why Attio is now the fastest-growing CRM vendor in Ramp's category data.
What is Attio MCP?
The Model Context Protocol is an open standard that lets AI assistants talk to external systems in a consistent way. Think of it as a universal adapter between an AI model and your tools. An MCP server exposes a set of actions, and any MCP-compatible client can call them.
Attio MCP is Attio's own hosted implementation of that standard. It gives AI tools secure access to your workspace and is built to work with Claude, ChatGPT, Cursor, and more. There is one endpoint to remember, https://mcp.attio.com/mcp, and authentication runs over OAuth, so there are no API keys to generate, store, or rotate.
This is the piece that turns Attio from a database into an operator surface. The REST API has always let developers build custom integrations. The MCP server does something different: it hands your live CRM to the AI tools your team already uses, with permissions that mirror each person's own access.
Attio MCP vs Ask Attio vs the API: which one you need
There are now three ways AI can touch your Attio data, and they solve different problems. People conflate them, so here is the clean split.
Access layer | Best for | How it works |
|---|---|---|
Ask Attio | Quick questions inside the app | Natural-language querying built into Attio |
REST API | Custom apps and data pipelines | Developers build integrations with API keys |
Attio MCP | Connecting external AI agents | One OAuth link exposes your CRM to Claude, ChatGPT, Cursor, and more |
Ask Attio is Attio's built-in assistant. It is fast for questions you ask while already looking at your CRM. Attio also ships an AI Research Agent that runs live web research inside workflows. Both live inside the Attio product.
The Attio MCP is the opposite direction of travel. Instead of bringing AI into Attio, it brings Attio into your AI. A common question is why you would do that when Attio already has its own assistant. The answer is orchestration. The value is not just conversational CRM, it is letting one agent work across Attio and every other MCP-connected tool at the same time. Claude can read a call in your notes, update the deal in Attio, and open a ticket in another system in a single run.
How to connect Attio MCP to Claude
Setup takes a couple of minutes. Attio lists only two prerequisites: an active Attio workspace and an MCP-compatible client. Here is the path for each client, based on Attio's setup docs.
Claude Desktop or Claude.ai. Open Settings, then Connectors, then Browse connectors. Search for Attio and install it. You will be sent through an OAuth login, and once you approve access, Claude is connected to your workspace. This is the one-click route and the one most GTM teams should use.
Claude Code. For the CLI, add the hosted server as a remote MCP connection and authenticate. Claude Code supports remote MCP servers, so the command is:
Then run /mcp inside Claude Code and complete the OAuth prompt. This is what powers the "attio mcp claude code" workflows people are building, where a coding agent reads a customer list and acts on it across tools.
ChatGPT. Open Apps, search for Attio, and connect. Attio is available natively in Claude, ChatGPT, and Notion.
Any other MCP client (Cursor, and others). Add https://mcp.attio.com/mcp as a remote MCP server and complete the OAuth flow when prompted. The same endpoint works everywhere, which is the point of the standard.
What the Attio MCP can actually do
The server exposes a broad set of tools grouped by object type. The real power comes from combining them in one prompt: search the workspace, create records, log notes, and update a pipeline in a single request. Here is the shape of it.
Category | What agents can do |
|---|---|
Records & objects | Search, create, and update people, companies, deals, and custom objects |
Lists | Read and update pipeline lists and entries, move deals between stages |
Notes & comments | Log notes, semantic-search past notes by topic, add comments |
Tasks | Create, list, and update follow-ups with deadlines and owners |
Meetings, calls & emails | Search recordings, transcripts, and emails, including semantic search |
Reporting & SQL | Run aggregate reports, and read-only SQL on supported plans |
A few tools deserve a callout because they change what is possible. semantic-search-notes and semantic-search-call-recordings let an agent find "notes where we discussed pricing objections" without exact keywords. run-basic-report produces grouped aggregates like open deals by stage. query-particle-sql runs read-only SQL for weighted pipeline math, though Attio notes it is not available on every plan. Write tools like create-record, upsert-record, and update-list-entry-by-record-id are what let the agent act, not just report.
Five GTM workflows worth running through Attio MCP
Setup is the easy part. Here is where the MCP actually moves revenue for a seed to Series B team. Each of these is a single prompt or a short chain.
Turn research into pipeline. Ask Claude to research target accounts in a segment, then create the company records in Attio, add them to a prospecting list, and draft first-touch emails. The whole top-of-funnel motion happens in one place instead of across four tabs. This is the workflow Attio highlights as the headline use case, and it is the fastest way to feel the difference.
Update the CRM straight after a call. Point your agent at a call recording or meeting note and have it update the deal stage, log a structured note, and create the follow-up task. If your notes live in Granola or Fireflies, the agent reads the summary and writes the outcome to the right record. The admin that used to pile up gets done in seconds.
Run pipeline hygiene on demand. Ask "which open deals have had no activity in the last 14 days" or "show weighted pipeline by stage and owner." The agent uses search, reporting, and SQL tools to answer in conversation. This is the review most teams skip because it is tedious, and now it is one prompt.
Orchestrate across your stack. The MCP is most powerful next to other MCP servers. Enrich new accounts with Clay, sync usage from your product, or push a flagged feature request from a deal into your engineering tracker. Your CRM context sits at the center while the agent works across tools.
Recall anything from your history. Semantic search over notes, calls, and emails means an agent can answer "what did we tell Acme about onboarding in Q3" by reading the actual transcript. Institutional memory becomes queryable instead of buried.
The pattern across all five is the same. The MCP does not just answer questions about your CRM, it does the work inside it.
An Attio MCP prompt library to start with
The fastest way to get value is to copy a prompt that already works and adapt it. These are organized by GTM motion, from top of funnel through reporting. Each one uses the tools covered above, and the agent asks you to confirm any write before it runs.
Outcome | Prompt to paste |
|---|---|
Build a target list | "Research the top 20 Series A fintech companies in the US and create a company record in Attio for each." |
Fill a prospecting list | "Add every company from that batch to the Q3 Outbound list and draft a first-touch email for each." |
Find decision makers | "Find contacts with a CTO or VP Engineering title at companies over 50 people." |
Update after a call | "Read my last call with Acme, log a note on the deal, move the stage to Proposal, and add a follow-up task for Friday." |
Clear the post-call backlog | "Summarize each of my calls this week and log a note on the matching account." |
Reconstruct an account | "What was our last interaction with Linear across notes, emails, and meetings?" |
Catch stalled deals | "List open deals with no activity in the last 14 days and tell me who owns each." |
Audit pipeline hygiene | "Which open deals are missing a next step or a close date?" |
Report by stage | "Count open deals by stage and show the total weighted value." |
Report by rep | "Show average deal size by owner this quarter." |
Segment the source mix | "Give me the source mix for deals over $50k." |
Handle objections | "Find calls where a prospect raised a pricing objection and summarize how we responded." |
Track competitors | "Search notes for any mention of our main competitor in the last 90 days." |
Deep-brief an account | "Pull everything we know about TheSwarm across records, notes, and emails." |
Enrich new records | "Enrich every company added this week with Clay and update headcount and funding." |
Route product feedback | "Turn the feature requests logged on open deals into a prioritized list for engineering." |
Keep prompts specific. Name the object, the list, and the outcome you want, and the agent picks the right tools. Vague prompts are where agents guess, and a clean data model is what keeps those guesses correct.
Security, permissions, and rate limits
Handing an AI agent write access to your CRM is a reasonable thing to be nervous about. Attio's model is built around that concern, and it is worth understanding before you connect.
Authentication runs over OAuth against your existing Attio login, so there are no API keys to leak, and you can revoke a session from your account settings at any time. Access is user-scoped: the agent sees exactly what you can see and can do only what you can do, no more. Read operations are auto-approved so search stays fast, while write operations request your confirmation before anything is created or changed, according to Attio's security documentation. Every operation is logged and auditable, and data stays inside your workspace.
The server also applies per-workspace rate limits, grouped into tiers. Normal conversational use stays well within them, but they matter if you script bulk operations.
Tier | Limit |
|---|---|
Read | 100 requests / second |
Write | 25 requests / second |
Search | 300 requests / minute |
Semantic search | 2 requests / second |
If an agent hits a limit, the fix is to reduce parallel operations and space out requests. For most GTM use, you will never notice the ceiling.
Troubleshooting the Attio MCP
Most connection problems come down to four things: the client, the login, the workspace, and permissions. Here is how to clear the common ones.
Attio does not appear in your client. Use the built-in connector directory first, which is Settings, then Connectors, then Browse connectors in Claude, and Apps in ChatGPT. If your client has no directory, add the remote server manually at https://mcp.attio.com/mcp and complete the OAuth prompt. Older app versions may not support connectors, so update the client first.
The OAuth login loops or fails. Make sure you are signed into the correct Attio account in the same browser, then retry. If it still fails, revoke the session from your Attio account settings and reconnect. You choose which workspace to authorize during login, so pick the right one if you belong to several.
The agent connects but finds nothing. The MCP authenticates as you, so it only sees what your Attio permissions allow. Confirm you connected the right workspace by asking "what workspace am I connected to," and if records are missing, have the agent list the object's attributes first so it queries the correct fields. Vague object names are a common cause of empty results.
Writes do not seem to happen. Write actions require your confirmation. If you dismissed the approval prompt, nothing was created or changed. Re-issue the request and approve it when the agent asks.
You hit rate-limit errors. Bulk jobs can trip the limits, especially semantic search, which is capped lowest. Reduce parallel operations, space out requests, and retry after a short pause. Normal conversational use stays well under the ceiling.
A tool is unavailable. Some tools are plan-dependent. Read-only SQL is the main one, so if the agent reports it cannot run SQL, check whether your Attio plan includes it.
Honest limitations
The MCP is powerful, and it is not magic. A few things are worth going in clear-eyed about.
It is only as good as your data model. If your objects are inconsistent, your stages are ambiguous, or half your fields are free text, the agent will reflect that mess back at you and sometimes guess wrong about where data belongs. The teams getting the most out of the MCP are the ones whose Attio schema was designed deliberately in the first place. If you are still building that foundation, our complete guide to Attio CRM covers the data model, and if you are moving off another tool, how to migrate from HubSpot to Attio keeps that schema clean on the way in.
Write actions still need judgment. Confirmation prompts protect you from silent damage, but an agent following a vague instruction can still create duplicate records or update the wrong entry. Clear prompts and a clean schema do most of the work here.
Some capability is plan-dependent. Read-only SQL, one of the more powerful reporting tools, is not on every billing plan. And the MCP is an access layer, not a replacement for well-built workflows and automations. Conversational actions are excellent for ad hoc work, but recurring, mission-critical processes still belong in Attio's workflow engine where they run without a human in the loop.
None of this is a knock on the product. It is the difference between plugging in a connector and building a system. Getting the data model and the workflows right up front is exactly what makes the MCP feel like a superpower instead of a party trick.
Frequently asked questions
How do I connect Attio to Claude?
In Claude Desktop or Claude.ai, open Settings, then Connectors, then Browse connectors, search for Attio, and install it. Approve the OAuth login and Claude is connected to your workspace. For Claude Code, add https://mcp.attio.com/mcp as a remote server and authenticate with the /mcp command.
Is the Attio MCP server free to use?
Attio does not sell the MCP as a separate add-on. You connect with your existing Attio account, and the listed prerequisites are just an active workspace and an MCP client. Some tools like read-only SQL are plan-limited, so check Attio's pricing and our pricing guide. Your AI subscription is billed separately.
Why use the Attio MCP instead of Attio's built-in AI?
Ask Attio answers questions inside the Attio app. The MCP brings your CRM into the tools where you already work, like Claude and ChatGPT, and lets one agent act across Attio and other connected systems at once. It is about orchestration and write actions, not just querying.
Can I use the Attio MCP with Claude Code?
Yes. Claude Code supports remote MCP servers, so you add Attio with claude mcp add --transport http attio https://mcp.attio.com/mcp, then run /mcp to complete the OAuth login. From there a coding agent can read lists, create records, and act on your CRM inside a larger automated workflow.
Is it safe to give an AI agent access to my Attio CRM?
The connection uses OAuth, not API keys, and access is scoped to your own Attio permissions. Read operations are auto-approved while writes ask for confirmation, sessions are revocable from your settings, and every action is logged. You control which workspace connects and can disconnect at any time.
Do I need a third-party Attio MCP server?
No. Attio now runs an official hosted server, which is the one to use for security and reliability since it authenticates over OAuth and is maintained by Attio. Community and directory options exist, but for production GTM work the official server at https://mcp.attio.com/mcp is the right default.
Sparsh Gupta, Founder of Automation Jinn and an Official Attio Expert Partner, helps seed to Series B B2B SaaS teams build AI-native GTM systems on Attio. If you want your CRM wired so agents can run real workflows without breaking your data, book a discovery call.
