ClaudeHubSpotMCPRevOps

Claude and HubSpot: why GTM teams need a working list, not just an MCP connection

Claude can read your HubSpot CRM and update records from the chat. A working list is where the rows, the checks and the hand-off stay for the next run.

ListPlus Team14 min read

A Claude HubSpot connection is HubSpot's connector for Claude, and it runs on HubSpot's MCP server. It is the right tool when someone asks Claude about a deal, a contact or last month's conversations, and when a change is small enough to confirm in the chat. A GTM, RevOps or sales-ops team that runs the same rows every week needs one more layer between the agent and the CRM: a working list that still holds those rows after the chat is closed, with checks, rules, a hand-off, a daily budget per connection and permissions per list.

As of 3 October 2026, The GTM Directory counted 456 of 973 GTM tools publishing an official MCP server, checked on each vendor's own pages (https://thegtmdirectory.com/research/state-of-mcp-in-gtm-tools-2026/2026-10). A connection is becoming an ordinary part of the stack. The list the team works from is the part that has to last from one run to the next.

What a Claude HubSpot connection does

HubSpot calls it the HubSpot connector for Claude and documents it for the Claude web, desktop and mobile apps. Its knowledge base guide, last updated 17 November 2026, describes it as a way to bring HubSpot context into Claude: Claude reads HubSpot records and can create or update them from the chat. The guide lists the connector for all HubSpot products and plans and asks for a paid Claude plan: Pro, Max, Team or Enterprise. HubSpot's developer docs add that its connectors for Claude, Microsoft Copilot, Gemini and ChatGPT are powered by the HubSpot MCP server (https://developers.hubspot.com/docs/build-with-ai/remote-mcp-server).

HubSpot MCP in the Claude app

HubSpot's setup is short. A Super Admin, or a user with App Marketplace Access, can connect without a prior approval. Everyone else waits until a Super Admin approves the connector, chooses which optional data it may use, and chooses who may install it.

  1. In Claude, open Settings, then Connectors, then Browse.
  2. Select HubSpot and choose Connect.
  3. Log in to the HubSpot account, choose the permissions to allow, then choose Connect app.
  4. In a chat, open + and Connectors, turn HubSpot on and ask the question.

One HubSpot account can be connected to a Claude account at a time. To disconnect, open Settings, then Connectors, choose HubSpot and Disconnect.

What it can write, and where it stops

HubSpot's table lists read, create and update for contacts, leads, companies, deals, tickets, custom objects and engagements: calls, meetings, notes, tasks and emails. Marketing assets such as landing pages, blog posts and marketing emails can be read, created and updated too. Conversations, users and teams are read-only. The connector does not delete, and it creates or updates 10 records at a time. Claude reaches only what the user's existing HubSpot permissions allow.

Conditional property rules and pipeline stage validation still apply; association label validations do not. HubSpot recommends setting the connector's write tools to Needs approval, because with Always allow, edits may happen without asking. The guide also says Claude can make mistakes and asks for a human review of what it produces. Every action is attributed in HubSpot's Audit Log to the user and to the Claude connector.

That fits Claude for sales work that is a question or a small edit: summarize open deals, pull the history before a call, log a note, create a follow-up task. HubSpot's page on AI tools names connectors for ChatGPT, Gemini and Microsoft Copilot as well (https://www.hubspot.com/ai-tools). For other MCP clients, HubSpot's developer docs describe the same server at https://mcp.hubspot.com, generally available to all HubSpot accounts since 13 April 2026 and signed in with OAuth through an MCP auth app you set up in HubSpot. The steps for that are in HubSpot's developer docs; this article does not write out a Claude Code setup HubSpot has not published.

What a repeatable list still needs

The connector answers the conversation in front of you. Close the chat and the next session starts from HubSpot again, unless the team has a place that kept the rows, the decision and what already happened. A CSV exported from that chat is one snapshot. It does not remember that a row was checked, that a work email was filled, or that the row already went across to the CRM.

A working list is that place, between the agent and the CRM. The agent — Claude, ChatGPT, Claude Code or n8n — runs where you run it. ListPlus does not run the agent. It is the list, and the tools the agent calls on it: search, checks, and enrichment of a missing field when a value is found. It is not a contact database you buy, and enrichment is one step on the list rather than the whole product.

What the working list keeps

Rows that are still there tomorrow

A row added to the list stays there, with its cells, its status and its tags. Adding records matches on email, LinkedIn profile, or name and company, and reports which rows were added, merged or rejected. The agent does not delete records. Hiding a row is as far as it goes. A hidden row leaves the working view, and the checks, rules, enrichment and sync skip it, until someone shows it again.

Rows can arrive from a file, from a webhook or a form on the list, from a HubSpot import the team runs in the app, or from a search the agent runs. A people search can be priced first, at 0 credits, and is charged per page it then loads. The rows land in a list in the workspace, with name, title, company and LinkedIn, and with no email or phone. Those come from a later lookup, charged when a value is found.

Checks before a credit is spent

The same checks the app runs are a tool the agent can call on a shared list, at 0 credits. The answer is a score, the findings, the affected records and a proposed fix. Nothing in the cells changes until a fix is applied, and applying one needs Edit on that list. A check the agent runs on JSON it sends, rather than on a list, is also 0 credits and stores nothing. Either way the next run can ask again, because the rows are still the same rows.

Rules that run on the next row too

A rule on the list is the part that repeats: when a record is new, changes, or meets conditions, set a status, add a tag, hide it, enrich it, or hand it to a pipeline that already exists. An enrich action names the fields, a ceiling per record, and whether it fills empty cells only. A step that would pass the ceiling is not started. A work email or a mobile number is charged when a value is found.

You can save that rule yourself. With the connection's Set up right ("May set up lists, columns and rules"), which stays off until you turn it on, the agent can create or change one too. A rule that sets a status, adds a tag or hides a record needs the list shared as Edit. A rule that enriches or hands records to a pipeline needs Full, and so does changing a rule someone else set up. An enrich rule the agent sets up can stay paused until you have agreed the budget, and it counts on that connection's day budget. Save the rule yourself and from then on it is yours, and it counts like any rule you saved. The agent does not create pipelines. You set a pipeline up once, and the agent hands records to it.

The hand-off back to HubSpot

A HubSpot pipeline matches a record by email. It creates a new contact, or treats one that is already there the way you chose: only fill empty fields (the default), leave it and only create new ones, or overwrite with your values. It can also add the contact to a static list in HubSpot. Only a contact the pipeline creates receives the ListPlus ID property, so a HubSpot workflow of your own can pick up the new contacts. A contact that was already in HubSpot never gets it, even when ListPlus fills its empty fields. ListPlus does not place contacts into a HubSpot sequence. A record without an email is held rather than sent.

If the pipeline is manual, it waits until a person releases the records, and the price is charged then. Handing rows to it costs the agent 0 credits. A pipeline that sends by itself charges per record it delivers: 2 credits for HubSpot, or 0.5 when the contact was already there and nothing was written. When the agent hands rows to such a pipeline, that price for those rows is reserved on the connection's day budget, and the hand-off is refused when the day has no room for it.

Importing from HubSpot into the list, and sending back, uses a connected HubSpot account. Writing needs that account connected with write access: ListPlus stops before it writes with a read-only account, and write access is granted under Settings → Connected accounts, on the same account. Pro includes every connected tool. On the Basic plan, one connected tool can be kept for 50 credits per 30 days. The agent itself can send rows only when that list is shared as Full. A rule you saved hands rows over on its own, including rows the agent only added.

A day budget on each connection

Every connection has its own day budget, in workspace credits. You set it when the agent signs in, and you can change it later. A new connection starts at 100 credits a day. Set it to 0 and the agent can only call actions that cost 0 credits: the checks, reading and updating lists you shared, a person lookup (no email or phone in the answer), a market count. Paid steps stop when that day's budget is used, including an enrich rule the connection set up and a hand-off to a pipeline that sends by itself. Rows waiting on such a rule stay waiting and continue after midnight UTC, or when you raise the budget. Revoking the connection stops it at once. A job still running is cancelled, and you pay for what it had already found.

Permissions on each list

An owner or an admin shares lists one by one. Nothing is shared until then. The default is Read + check: the agent can read records and run the checks, and it writes nothing. Edit also adds records, changes cells, sets statuses and tags, hides rows and applies fixes. Full also fills missing fields, fills AI columns and hands records to pipelines. Under that preset you can hide columns the agent should never see, choose the columns it may write, let it fill only empty cells, and limit it to some sublists. Lists the agent creates for a search are shared with that connection as Full, and you can narrow them the same way.

Claude Code, ChatGPT and n8n use the same list

Claude, ChatGPT and Claude Code sign in with the ListPlus account. Any other client, including n8n, Make, Zapier or your own code, sends a key and calls the same actions over REST at https://listplus.ai/a/v1/{action}. The MCP address is https://listplus.ai/mcp. Connecting the agent is included on every plan, including Basic.

In the Claude apps, ListPlus is a custom connector: Add to Claude, on the Your AI page in ListPlus, opens Claude's Add custom connector dialog with the name and address filled in. Anthropic offers custom connectors on every Claude plan: Pro, Max, Ultra, Team and Enterprise, and the plan without a subscription, which is limited to one custom connector. On Team and Enterprise an admin adds it in the organization settings, and each member then connects (https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp). For Claude Code the command is:

claude mcp add --transport http listplus https://listplus.ai/mcp --scope user

Then sign in with /mcp. Claude Code and the CRM stay in two places: Claude Code works the list through that connection, and HubSpot stays the system the pipeline writes to. ChatGPT adds the same MCP address as a plugin and signs in with OAuth. n8n uses one HTTP request per action, with the key in the header. The prices and the permissions belong to the connection, whichever client calls it.

A workflow you can run again

Nothing below is a result from a customer account. It is the path the product runs.

  1. In the Claude app, connect HubSpot and use it for questions about deals, contacts and history, and for a change you review before it is written.
  2. Connect that Claude, or Claude Code, ChatGPT or n8n, to ListPlus and set the day budget.
  3. Share one list. Put rows in it from a file, from a HubSpot import, or from a people search the agent runs. The search result has no email or phone yet.
  4. Ask the agent to run the checks. Apply a fix only where you agree, which needs Edit. Hide what should leave the working view.
  5. Save a rule: rows that fit are enriched for a missing work email, at most a ceiling you set per record, empty cells only. The lookup is charged when an address is found.
  6. Add the HubSpot pipeline you already set up as the next action of that rule, or send the qualifying rows once the list is shared as Full. The next run opens the same list: who was checked, which cells were filled, which rows already went across.

When the CRM is Salesforce

A Salesforce MCP connector, where your team uses one to read Salesforce from Claude, is the chat side of the same split. This article does not describe that connector. On the list side, a pipeline you set up can hand the rows to Salesforce as a lead or a contact. A new lead needs a last name and a company, a new contact a last name, and a record without an email is held. The checks, the rule and the day budget still sit on the list.

Where to start

Use the HubSpot connector when the job is a question or a small, confirmed edit inside Claude. Use a working list when the job is a set of rows you will check, enrich only where a value is missing, and hand to the CRM again next week. ListPlus is that list for the agent you already run. Checks cost 0 credits. A HubSpot pipeline charges 2 credits per contact it delivers. A page of a people search is charged when it is loaded. A missing work email or mobile is charged when it is found.

Questions before you connect

How do I connect Claude to HubSpot?

In the Claude web, desktop or mobile app: Settings, then Connectors, then Browse. Select HubSpot, choose Connect, log in to HubSpot, choose the permissions to allow and confirm with Connect app. In a chat, turn HubSpot on under + and Connectors. A Super Admin, or a user with App Marketplace Access, connects without a prior approval; everyone else waits until a Super Admin approves the connector. HubSpot's guide, last updated 17 November 2026, lists the connector for all HubSpot products and plans and asks for a paid Claude plan: Pro, Max, Team or Enterprise.

Can Claude write to HubSpot, and is it safe?

It can create and update, within limits HubSpot publishes. The connector reads, creates and updates contacts, leads, companies, deals, tickets, custom objects and engagements; conversations, users and teams are read-only. It does not delete, and it creates or updates 10 records at a time. Claude reaches only what the user's own HubSpot permissions allow, and every action is attributed in the Audit Log to the user and to the connector. HubSpot recommends setting the write tools to Needs approval; with Always allow, edits may happen without asking.

What is a HubSpot MCP connection good for?

Reading CRM context in the chat, and making a change you confirm there. A larger set of rows, with checks and a rule you will run again, belongs on a working list: the connector creates or updates 10 records at a time, and the next chat does not remember which rows were already checked or sent.

What does it cost?

HubSpot's guide names no price for the connector; the Claude side needs a paid plan. On ListPlus, connecting an agent is included on every plan, including Basic, and the checks cost 0 credits. A HubSpot pipeline that sends by itself charges 2 credits per contact it delivers, or 0.5 when the contact was already there and nothing was written. A missing work email or mobile is charged when a value is found. Importing from HubSpot and sending back need a connected HubSpot account: included in Pro, or on Basic as the one connected tool for 50 credits per 30 days.

Does ListPlus run Claude, ChatGPT or Claude Code?

No. The agent runs where you run it. ListPlus is the working list and the tools that agent calls over MCP or REST: search, checks, and enrichment of a missing field when a value is found. You choose which lists it may touch and the day budget on the connection.

What does the daily budget on a connection cover?

Workspace credits, up to the budget on that connection. A new connection starts at 100 credits a day, and you can change it. Set it to 0 and the agent can only call actions that cost 0 credits. An enrich rule that connection set up counts on the same day budget, and so does a hand-off to a pipeline that sends by itself. When the day is used up, the rule's rows wait and continue after midnight UTC, or when you raise the budget; a hand-off that does not fit is refused.

Which lists can the agent see, and what can it change?

Only lists a workspace owner or admin shares with it, each as Read + check, Edit or Full, plus lists the agent created itself. Read + check is the default and writes nothing. You can hide columns, limit writes to empty cells, and limit the agent to some sublists. Records are not deleted through an agent. Hiding a row is as far as that goes.