How to Connect Linear to ChatGPT (Step-by-Step)

Linear holds the actual state of what your team is building — every bug, every cycle, every half-finished project. Connecting it to ChatGPT lets you ask questions across all of it instead of building yet another filtered view: what is blocked, what shipped, what keeps coming back.

This guide covers the setup, the invocation step that trips almost everyone up, the tool-call ceiling that makes broad requests fail, and what the connector genuinely cannot do.

What You Need Before Starting

• A paid ChatGPT plan — connectors are not available on the free tier
• A Linear account with access to the workspace you want to query
• Admin approval on Business or Enterprise — an administrator may need to enable the connector first

Step 1: Connect Linear in Settings

Click your profile icon, then Settings, then Apps & connectors. Find Linear and click Connect. You are redirected to Linear to sign in and approve access for your workspace, then returned to ChatGPT.

If Linear is not in the list, the usual explanation on a Business or Enterprise plan is that an administrator has not enabled it for your workspace — not a problem with your account.

Step 2: Actually Invoke It (Where Everyone Gets Stuck)

This is the part worth reading carefully. Connecting Linear does not make ChatGPT use it. Ask a question in a normal chat and ChatGPT will answer from general knowledge, cheerfully and without touching your issues at all.

To query Linear, click Tools in the message box, choose Use connectors or Run deep research, select Linear as a source, then type your prompt. Nearly every “the Linear connector does not work” report comes down to this step being skipped.

Step 3: Ask Questions Across Your Issues

Once Linear is selected as a source, ask for the answer rather than for a filter. Questions that work well:

• “Summarise every open bug tagged billing and group them by severity.”
• “What shipped in the last two cycles that touched onboarding?”
• “Which issues have been reopened more than once, and what do they have in common?”
• “Draft a cycle retrospective from what actually closed versus what we planned.”

The last one is where this earns its keep. Pattern questions across many issues are exactly what a saved view cannot answer and what a person would otherwise spend an afternoon on.

Step 4: Keep Requests Scoped (the Tool-Call Ceiling)

Connector sessions have a practical limit on how many tool calls the model will make before it stops. Users have reported the agent giving up after roughly three to five calls on broad requests, which reads as the connector failing when it is really a budget being exhausted.

Work with it rather than against it: one well-scoped question per turn. “Summarise open billing bugs” will finish. “Audit the entire backlog across every project and tell me what to prioritise” will not. If you need the sweep, break it into several prompts and combine the answers yourself.

What the Connector Cannot Do

Two limits worth setting expectations on before you roll this out to a team:

• It reads, it does not write. The connector will not create or update issues. Writing back into Linear is an API automation project, not a setting.
• It is not Codex in Linear. Those are different products that get confused constantly. The connector answers questions about your issues; Codex in Linear is a coding agent that picks up a ticket and produces code. Planning and reporting is the connector; implementing a ticket is Codex.

Connectors also behave differently inside ChatGPT Projects, where users have reported them not appearing. If Linear is missing from a Project, test the same query in a regular chat before troubleshooting the connection itself.

The Canadian Angle: What to Check Before Connecting

On Business and Enterprise plans, OpenAI states that your data is not used to train its models. Beyond that, the standard Canadian considerations apply: OpenAI processes data in the United States, so a PIPEDA-covered organization should disclose the cross-border transfer, assess the vendor and confirm contractual safeguards, and Quebec's Law 25 requires a privacy impact assessment before personal information is communicated outside the province.

One point specific to issue trackers: they usually hold more personal information than teams expect. Customer names and email addresses in bug reports, staff performance discussions in comments, security findings in tickets. Have a look at what is actually in there before connecting — our AI consulting practice runs these reviews for Canadian organizations, and the Business vs Enterprise comparison covers which tier gives admins the controls to enforce a decision.

Step 5: Manage or Revoke Access

Go to Settings → Apps & connectors and click Manage next to Linear to disconnect. Revoking takes effect immediately. Add it to your offboarding checklist — a connector authorised by someone who has since left is an access path nobody is reviewing.

Frequently Asked Questions