Linear MCP with Claude Code: turn Linkeddit findings into issues

Linear hosts its MCP server at https://mcp.linear.app/mcp and says the setup flow uses OAuth 2.1. For Claude Code, run claude mcp add --transport http linear-server https://mcp.linear.app/mcp, then open a session and run /mcp to authenticate. In Claude on the web or desktop, Linear's docs point to the connectors page instead.

Linear's own Claude page warns that remote MCP connections are still early and may fail or need several attempts. Restarting the client, or disabling and re-enabling the server, is its suggested fix. If you work in more than one workspace, its docs say each workspace needs its own authentication context.

For a safer first run, Linear offers a read-only endpoint at https://mcp.linear.app/mcp/readonly, which only exposes read tools. Use it while you check what Claude would file, then switch to the standard endpoint when you want issues created.

The Linear connector listing shows 22 tools: 17 reads and 5 writes. The writes are create_issue, update_issue, create_comment, create_project and update_project. The reads include list_teams, list_issues, list_issue_labels and list_issue_statuses. It files and updates work, but it knows nothing about where AI engines recommend a competitor or what your competitors shipped this week.

Linkeddit supplies that evidence over its own connector at https://mcp.linkeddit.com/mcp: Answer Radar gaps and actions, Compete signals and page changes, and Search Console queries. It writes no advice. Claude reads the evidence, drafts the issues, and Linear receives only what you approve. Both connectors sit behind one Linkeddit plan at $60 a month.

Linkeddit's compete_signals_get returns competitor moves with a priority, a category, why it matters and a source link, filterable by competitor and week. On Linkeddit's test project, one call returned two signals for a tracked analytics competitor, both dated to the week of 2026-10-26, both priority low, both blog posts about its upcoming release. Most weeks are quiet. That is the point of filtering.

Ask Claude to keep only priority high or medium signals, check Linear with list_issues so the same move is not filed twice, and draft one issue per move with the source link in the description. Page changes work the same way: the page_changes view returns snapshots with a content hash, and a changed hash between two snapshots of one page marks a change.

Review the drafts before Claude calls create_issue. A competitor's pricing change belongs with the product owner, and a launch belongs with whoever writes positioning, so tell Claude the team and label for each category. Linear's own example prompts follow the same habit: they ask the assistant to show what it plans to create before it creates anything.

Linkeddit's gsc_queries_get reads your own Google Search Console data and has views for striking_distance (average position 11 to 20 with real impressions), ctr_opportunities (top 5 positions clicked far less than expected), no_dedicated_page and long_tail. Each is a ready list of pages worth a refresh, which is what a ticket needs.

Ask Claude to pull striking_distance, group queries by the page they land on, and draft one Linear issue per page: the queries, their impressions and average position, and what to change. Search Console must be connected first. On Linkeddit's test project it is not, and the tool said so plainly instead of returning an empty list, so connect the property before you expect rows.

Keep the batch small. A handful of refresh issues a week is easier to ship than a backlog of fifty, and the same view will show whether a page moved when you pull it again a few weeks later.

The first guide is AI visibility tasks: Answer Radar's action queue, filed as Linear issues with the evidence attached, and the cards moved forward as issues are done. It is live and is the best place to start.

Competitor alerts and Search Console refresh tickets are covered in this hub rather than in separate guides, because searches for them are rare and one page serves them better. Other Linear workflows are not written yet.