On June 23, 2026, HubSpot moved its Agent CLI from private to public beta — a command-line tool built specifically for AI agents like Claude Code, Claude Cowork, and OpenAI Codex to read, write, and manage HubSpot CRM data without a human clicking through a single dashboard. It’s not a chatbot, not another AI Connector, and not a replacement for HubSpot’s existing MCP server. It’s infrastructure for a different kind of work: the bulk, scheduled, unattended CRM housekeeping that no rep or ops manager has time to do by hand.

Key Takeaways

• HubSpot’s Agent CLI reached public beta on June 23, 2026, following a private beta that opened in late May 2026.
• It’s a single, dependency-free binary for macOS, Linux, and Windows that authenticates via OAuth or a service key and talks directly to HubSpot’s API.
• It’s purpose-built for headless, scheduled, and bulk agent workflows — not conversational, human-in-the-loop assistance, which HubSpot’s MCP server and AI Connectors already handle.
• Supported agent environments include Claude Code, Claude Cowork, and OpenAI Codex, with JSONL as the default output format for easy piping between tools.
• Every write command supports --dry-run, and Super Admins can gate access through HubSpot’s App Governance beta.
• HubSpot’s stated ambition, per Chief Product and Technology Officer Duncan Lennox, is for agents to both “run on HubSpot” and “run HubSpot” — a materially different posture than Salesforce and Microsoft’s tighter agent-access models.
• The shift raises real governance questions: agents with CRUD access to production CRM data need guardrails, not just capability.

What the HubSpot Agent CLI Actually Is

Strip away the branding and the Agent CLI is exactly what it sounds like: a command-line tool that gives an AI agent the same kind of structured access to HubSpot that a developer would get from the REST API, but packaged for how coding agents actually operate — reading instructions from a repo, running headless in a terminal, and executing multi-step plans without a chat window open.

According to HubSpot’s own documentation and changelog, the CLI supports the full range of CRM object operations — list, get, search, create, update, upsert, merge, and delete — across contacts, companies, deals, tickets, and custom objects. It also handles pipeline and pipeline-stage management, custom properties, associations between records, activity history, and workflow automation. Output defaults to JSONL, a line-by-line JSON format built for piping into other tools and scripts, with optional JSON-array or human-readable table formats available.

Installation is a single command — curl -fsSL https://api.hubapi.com/hub/cli/backend/hub-cli/latest/install.sh | sh on macOS/Linux, or the PowerShell equivalent on Windows — and the tool ships as a dependency-free binary. Authentication runs through either OAuth login (recommended, user-scoped) or a service key for elevated “admin mode” operations.

Why HubSpot Built a CLI When It Already Has an MCP Server

This is the part worth understanding clearly, because the distinction is the whole story. HubSpot already has an MCP (Model Context Protocol) server that connects AI assistants to CRM data — that’s the tool powering conversational use cases where a human is in the loop, asking Claude or another assistant to “pull my top 10 stale deals” and reviewing the answer in real time.

The Agent CLI is deliberately built for the opposite scenario: unattended, scheduled, bulk work. HubSpot’s own framing draws the line directly — AI Connectors and MCP excel when a human is present for conversation and analysis; the CLI exists for automation that runs in the background, on a schedule, without anyone watching. Think nightly data-hygiene sweeps, weekly pipeline-health reports generated before the team arrives, or batch enrichment jobs touching thousands of records — the kind of repetitive operational work that’s a poor fit for a chat interface but a good fit for a script an agent runs at 3 a.m.

HubSpot Chief Product and Technology Officer Duncan Lennox summed up the underlying philosophy bluntly: “Agents don’t click through dashboards or navigate interfaces; they call APIs, read structured outputs, and take action.” The company’s stated goal is for agents to both run on HubSpot — building custom tools and workflows against it — and run HubSpot itself, operating platform functions end-to-end.

Why This Matters for CRM Buyers, Not Just Developers

If you run a CRM instance for a mid-market business, this isn’t an abstract developer-tools story. It’s a signal about where CRM administration is headed. For years, “CRM automation” meant workflow builders, native automation rules, and third-party iPaaS tools like Zapier or Workato sitting between systems. The Agent CLI represents a fourth path: a general-purpose AI agent, running in a coding environment your team already uses for other work, that can be pointed directly at your CRM and told, in plain language, to go do something.

That has real implications for how CRM Experts Online’s clients — mostly mid-market companies running Zoho, Salesforce, HubSpot, or NetSuite — think about staffing operational work. A task like “review every deal that’s had no activity in 21 days, check if the primary contact still has a valid title, and flag anything inconsistent” used to require either a dedicated ops analyst or a custom-built integration. Now it’s a prompt an agent can execute against a CLI, on a schedule, with a dry-run preview before anything touches production data.

Practical Use Cases Already Emerging

Based on HubSpot’s own published examples and the broader pattern of what early adopters are building, the realistic near-term use cases cluster around a few categories:

• Data hygiene at scale. Weekly scans identifying incomplete or duplicate contact records, then either flagging them for a human or applying agreed-upon fixes automatically.
• Pipeline monitoring without a dashboard. Daily checks for deals with no recent activity, stalled stages, or missing close dates, surfaced as a summary before the sales team’s morning standup.
• Account reviews for CS and support teams. Pulling deal history, support ticket volume, and NPS scores into a single prepared summary ahead of a renewal or QBR — work that currently eats hours of a CSM’s week.
• Ticket triage and pattern detection. Automatically grouping and escalating tickets from top-tier accounts that show a recurring issue pattern.
• Bulk record operations. Merging duplicate companies, batch-updating properties after a rebrand or territory change, or backfilling custom fields from an external data source — the kind of one-time migration work that used to require a developer or a support ticket to HubSpot.

Benefits and Real Risks

The upside is straightforward: less manual, repetitive CRM maintenance work, faster turnaround on operational reporting, and a lower barrier to building custom automation without waiting on a developer sprint. For lean ops teams, that’s meaningful.

The risk side deserves equal weight, and HubSpot’s own documentation reflects that it’s taking this seriously. The CLI’s write commands operate directly on live CRM data — a misconfigured or overly broad prompt could merge the wrong records, delete active deals, or bulk-update the wrong property across thousands of contacts before anyone notices. That’s not a hypothetical: the broader industry has already seen incidents where an autonomous agent with unsupervised database access caused catastrophic, fast-moving damage. It’s exactly why HubSpot ships every write command with a --dry-run flag and recommends testing changes in a sandbox account first.

There’s also a governance dimension. HubSpot’s approach — pushing toward full API parity and letting agents “run HubSpot” — is a more permissive stance than Salesforce or Microsoft currently take with their own agent platforms, both of which maintain tighter control over what autonomous systems can touch. That’s a genuine trade-off: more automation capability paired with more exposure, and the tooling to manage that exposure (permission scopes, audit trails, approval workflows) has to mature at the same pace as the capability itself, or the risk outpaces the benefit.

Implementation Best Practices — and Common Mistakes

Businesses experimenting with the Agent CLI, or any agent-driven CRM automation, should treat it with the same discipline as a production API integration, because that’s what it is:

• Start with OAuth, user-scoped access — not the service key. Admin mode with a service key should be reserved for narrowly scoped, reviewed automations, not general experimentation.
• Use --dry-run on every new automation before it touches real data. This is the single most important habit to build, and it’s the one teams skip when they’re in a hurry.
• Gate access through App Governance. Super Admins can restrict who is allowed to connect the CLI at all — use it, especially in accounts where sensitive customer or financial data lives in custom objects.
• Test in a sandbox account first, especially for anything involving merge, delete, or bulk upsert operations.
• Don’t confuse “can automate” with “should automate unsupervised.” Scheduled agent runs should still report their actions somewhere a human reviews — a Slack digest, a summary email, an audit log — rather than running silently with no visibility.
• Common mistake: treating this as a drop-in replacement for a real integration strategy. The CLI is excellent for internal ops automation; it’s not a substitute for a properly architected system-to-system integration when reliability and error-handling guarantees matter (e.g., syncing to a finance system).

CRM Experts Online’s Perspective

We’ve spent years building custom automations for clients who needed exactly this kind of unattended operational work done — the reporting nobody has time for, the data cleanup that keeps slipping, the account reviews that get rushed together the morning of a renewal call. Historically, that meant either a developer writing custom API calls or a workflow builder stretched past its comfort zone with brittle conditional logic.

What the Agent CLI changes is the speed of that first build. An agent that can read HubSpot’s own API surface and iterate against a real account in a sandbox can prototype an automation in an afternoon that used to take a sprint. That’s genuinely useful — but it doesn’t remove the need for someone who understands the data model, the governance implications, and where a “helpful” automation can quietly cause damage at scale. We’re already testing the CLI against client sandbox instances specifically to build a playbook: which use cases are safe to hand to a scheduled agent today, and which still need a human checkpoint. If your team is evaluating whether to let an AI agent touch production CRM data — via HubSpot’s CLI, Zoho’s MCP tooling, or NetSuite’s AI Connector Service — that’s a conversation worth having with an implementation partner before you flip it on, not after something breaks.

FAQ

Is the HubSpot Agent CLI available to all HubSpot customers? It’s in public beta as of June 23, 2026, meaning any HubSpot account holder can access it, though beta status means commands and behavior may still change without notice.

Does the Agent CLI replace HubSpot’s MCP server or AI Connectors? No. HubSpot explicitly positions them as complementary — MCP and AI Connectors for conversational, human-in-the-loop work; the CLI for unattended, scheduled, bulk operations.

Which AI agents work with the Agent CLI? HubSpot’s documentation names Claude Code, Claude Cowork, and OpenAI Codex as supported environments, along with other compatible AI coding agents.

Can the CLI delete or modify live CRM records? Yes — it supports create, update, upsert, merge, and delete across CRM objects, which is exactly why HubSpot recommends sandbox testing and the built-in --dry-run flag before running any write operation against production data.

How do we control who can use it in our HubSpot account? Super Admins can restrict CLI connection permissions through HubSpot’s App Governance (beta) feature, requiring admin approval before a user can connect it via OAuth.

Is this safe for a business without an internal developer? It can be used without deep engineering resources, but the risk of a misconfigured bulk operation on production data is real. We’d recommend starting with read-only and sandboxed use cases before granting write access.

How does this compare to what Salesforce or Zoho are doing? HubSpot’s approach is notably more open — pushing toward full API parity so agents can operate the platform end-to-end — compared to the tighter agent-access controls Salesforce and Microsoft currently maintain.

What’s the cost? HubSpot has not published separate pricing for the Agent CLI as a distinct line item during the beta; it operates against your existing account’s API access.

Conclusion

The Agent CLI is a small tool with a large implication: CRM platforms are starting to build infrastructure specifically for autonomous agents, not just for human users or human-supervised chat assistants. That’s a meaningful shift in how operational CRM work gets done, and it’s moving fast enough that waiting to understand it isn’t a neutral choice. If you’re curious whether agent-driven automation makes sense for your HubSpot, Salesforce, Zoho, or NetSuite environment — and where the governance guardrails need to go — schedule a consultation with CRM Experts Online and we’ll help you figure out what to automate first, and what to leave to a human for now.

Further Reading

• Introducing the HubSpot Agent CLI — HubSpot Blog
• HubSpot Agent CLI available in public beta — HubSpot Developer Changelog
• Use the HubSpot Agent CLI — HubSpot Knowledge Base
• HubSpot/agent-cli — Official GitHub Repository
• HubSpot Prepares to Hand Over CRM Control to AI Agents — CX Today