Gmail Linear Exports

Gmail → Linear Exports

Fetch emails via the Gmail MCP, produce a privacy-safe digest with prioritized action points, confirm the target Linear team, then create one anonymized Linear issue per email/action point.

Prerequisites

• Gmail MCP connected and authenticated (mcp__claude_ai_Gmail__*).
• Linear MCP connected and authenticated (mcp__linear__*). This project's .mcp.json already defines the linear server; if its tools aren't loaded, tell the user to run /mcp → authenticate in an interactive session and stop.
• Both are remote OAuth servers — a non-interactive session cannot run the login flow.

Input

$ARGUMENTS (all optional) may narrow the scan:

• A date or range (default: today in the user's local timezone; use currentDate from context if present).
• A Gmail filter to append (e.g. is:unread, label:..., category:primary).
• A target Linear team name (skips the team question in Step 5).

Default with no arguments: today's inbox → the Testpawelhuryn-style default team after confirming.

Steps

1. Scan Gmail

Call mcp__claude_ai_Gmail__search_threads with:

• query: after:YYYY/MM/DD before:YYYY/MM/DD in:inbox (upper bound is the day after the target day; append any $ARGUMENTS filter).
• pageSize: 50.

Only keep messages whose date falls in the target window. A returned thread may include older messages (e.g. an old SENT message in a reply thread) — ignore those for counting/export.

2. Read what needs action

For each thread that looks like it needs a decision or action, call mcp__claude_ai_Gmail__get_thread with messageFormat: FULL_CONTENT to read the real body (snippets truncate and hide the ask). Skip full fetches for obvious noise (newsletters, automated notifications) — snippet is enough to classify.

3. Anonymize (hard rule)

Never echo — in the summary or in Linear — any of:

• Sender/recipient names or email addresses.
• Company, product, or platform names that identify a party.
• Verbatim subject lines, or names appearing inside snippets/bodies.

Refer to parties generically: "a vendor's security team", "a cold outreach", "a paid-collaboration pitch". Describe subjects by topic, not literal text. Keep the actionable substance (what's being asked, the next step) — strip only the identifying details.

4. Summarize for the user first

Present a tight anonymized digest before writing anything to Linear:

• One short line per email/action point.
• Flag which items have a genuine next action vs. which are decline/ignore.

5. Confirm the Linear target

Call mcp__linear__list_teams.

• If $ARGUMENTS named a team, use it.
• If exactly one team exists, use it.
• If multiple exist and none was specified, ask the user which team (a cheap, non-obvious choice) before creating anything.

6. Export to Linear

Create one issue per email/action point with mcp__linear__save_issue:

• team: the confirmed team.
• title: short, action-oriented, anonymized (e.g. "Resubmit security vulnerability report via bug-bounty platform").
• description: start with Source: inbox action point (anonymized)., then 1–2 sentences of anonymized context, then a Next step: line. Use literal newlines, not \n.
• priority by urgency:
  • 1 Urgent — time-critical, tied to a deadline today/tomorrow.
  • 2 High — a real response is expected and matters.
  • 3 Medium — a genuine next action, not time-critical.
  • 4 Low — decline / archive / FYI, no action really required.
• Leave state at default (Backlog) unless the user asked otherwise.

Send independent save_issue calls in parallel in one message.

7. Report back

Show a compact table: issue identifier (linked to its url), priority, title. Note how many have a real next action. Offer to set states/due dates or move issues if wanted.

Notes

• Skip self-sent / automated mail for export unless it carries a real action.
• Re-runs: this creates fresh issues each run; if the user re-scans the same day, ask before duplicating, or mention that duplicates will be created.
• Keep the whole chat response tight — this skill's value is a clean digest plus a clean set of issues, not a wall of text.