HubSpot has launched the Agent CLI in public beta, a command-line tool that lets AI coding agents like Claude Code, Claude Cowork, and OpenAI Codex read, create, update, and manage records directly in your HubSpot CRM. For teams already using AI agents for coding and research, this closes the gap between "AI that writes code" and "AI that can act on your customer data."
This matters because most AI-in-CRM conversations so far have focused on chat assistants inside HubSpot itself. The Agent CLI takes a different approach: it brings HubSpot data into the coding agent workspace your technical team already uses, so agents can run scheduled, bulk, or headless operations against CRM records — not just answer questions about them.
For RevOps, marketing operations, and CRM admins, the Agent CLI is a new automation surface worth evaluating carefully, especially around permissions and data governance. This post explains what it is, how it works, where it fits next to HubSpot's existing MCP server and APIs, and what to check before connecting it to a production account.
The HubSpot Agent CLI is a command-line tool, released in public beta in mid-2026, that connects supported AI coding agents to a HubSpot account. Once installed and authenticated, an agent such as Claude Code or OpenAI Codex can use the CLI to interact with contacts, companies, deals, tickets, custom objects, workflows, pipelines, properties, associations, and activity history — using the permissions of the connected HubSpot user.
Unlike chat-based AI features built into the HubSpot UI, the Agent CLI is designed for developers and technical operations staff who already work inside coding agent environments. It is installed with a shell command, authenticated through standard HubSpot OAuth, and paired with a skills library (hubspot/agent-cli-skills) that gives agents predefined guidance for CRM searches, bulk operations, data quality tasks, workflows, and reporting.
Most HubSpot AI discussion in 2026 has centered on in-app tools like Breeze and chat-based copilots. The Agent CLI represents a different automation pattern that matters for a few reasons:
At a high level, connecting and using the Agent CLI follows this sequence:
hubspot auth login, which opens a standard HubSpot OAuth flow.npx skills add hubspot/agent-cli-skills so the agent has predefined guidance for common CRM operations.--dry-run option where available for actions that modify or delete data.HubSpot's own guidance is explicit that generated recommendations and large-scale updates should be reviewed before being applied — the CLI is designed to assist technical staff, not to run unsupervised against production data by default.
| Team | Example Task | What the Agent Does |
|---|---|---|
| Marketing | Contact cleanup and re-engagement | Finds high-fit contacts with no deal or recent activity, flags missing enrichment fields |
| Sales / RevOps | Pipeline health monitoring | Checks deals closing soon with no recent activity and sends a summary |
| Sales leadership | Call transcript analysis | Summarizes talk time, competitor mentions, and hesitation themes across recent calls |
| RevOps / CRM admins | Property cleanup | Finds duplicate or deprecated properties, previews impact before removing them |
| Operations | Ticket consolidation | Merges duplicate support tickets while preserving source details |
| Customer Success | Account review prep | Summarizes open deals, support activity, and NPS score per account |
HubSpot now offers several ways for AI to interact with CRM data. Choosing the right one depends on who is using it and what kind of task is involved.
| Access Layer | Best For | Interaction Style | Governance Considerations |
|---|---|---|---|
| Agent CLI | Developers and technical ops running scheduled/bulk tasks | Command-line, scriptable, skill-based | Requires App Install Governance review; acts on connected user's permissions |
| HubSpot MCP Server | Connecting general-purpose AI assistants (e.g., Claude Desktop) to HubSpot | Conversational, tool-call based | Similar permission scoping; more suited to ad hoc queries than scheduled jobs |
| HubSpot REST APIs | Custom integrations and applications built by developers | Direct API calls, full control | Requires custom code and ongoing maintenance |
| Native HubSpot AI (Breeze, in-app copilots) | Business users working inside the HubSpot UI | Point-and-click, chat-based | Governed by existing HubSpot user permissions and seat licensing |
Choose the Agent CLI if your team already uses Claude Code, Cowork, or Codex for engineering work and wants to extend that same workflow to recurring CRM operations.
Choose the MCP server if you want a general AI assistant to answer questions about CRM data conversationally, without scripting scheduled jobs.
Choose direct API integration if you need tightly controlled, purpose-built automation that doesn't depend on a third-party coding agent.
Before connecting the Agent CLI to a production HubSpot account:
hubspot/agent-cli-skills) is maintained externally and should be reviewed periodically with npx skills update.Vantage Point helps organizations evaluate and govern new HubSpot automation surfaces like the Agent CLI as part of broader CRM strategy. That includes reviewing permission models before rollout, mapping which HubSpot access layer (Agent CLI, MCP, API, or native AI) fits a given use case, and building workflows that use AI safely without compromising data quality.
If your team is evaluating how this applies to HubSpot, CRM governance, or automation strategy, Vantage Point can help assess the right next step and build a practical implementation plan. Our HubSpot services cover platform optimization and AI enablement, and our CRM and marketing automation work extends into workflow design once new tools like this are connected. For teams running both platforms, we also support HubSpot and Salesforce integration so CRM data stays consistent regardless of which AI access layer touches it.
The HubSpot Agent CLI is a public beta command-line tool that connects AI coding agents such as Claude Code, Claude Cowork, and OpenAI Codex to a HubSpot account, allowing them to read and modify CRM records, workflows, and properties.
HubSpot's documentation lists Claude Code, Claude Cowork, and OpenAI Codex as supported agents, along with other compatible AI coding agents that can run command-line tools.
It can be, but HubSpot recommends starting with reporting and analysis tasks, reviewing proposed changes before applying them, and using the --dry-run option for actions that modify or delete data. Super Admins should also review App Install Governance settings before allowing broad access.
It is best suited to RevOps teams, CRM administrators, and technical marketing or sales operations staff who already work in AI coding agent environments and want to automate recurring CRM tasks like cleanup, reporting, or pipeline review.
The MCP server is designed for conversational AI assistants asking ad hoc questions about CRM data, while the Agent CLI is built for scriptable, scheduled, or bulk operations run from a coding agent workspace.
Yes. The CLI operates using the permissions of the connected HubSpot user, which can include creating, updating, or deleting records. This is why HubSpot recommends reviewing proposed changes and using dry-run previews before applying updates.
Business users can prompt an already-connected agent with plain-language instructions, but installation, authentication, and permission review typically require a developer or CRM admin familiar with command-line tools and HubSpot's permission model.
Not necessarily. For tightly controlled, purpose-built automation, a direct API integration may still be the better fit. The Agent CLI is most useful for recurring, agent-driven tasks that don't warrant custom integration code.