MCP — recipes

These are example prompts that combine multiple MCP tools to get something useful done. Each one assumes your AI tool has the bichito MCP server installed (see Install).

Triage

Show me my open critical bichitos and resolve the ones from honeycomb X

"List my critical open bichitos. For any that belong to honeycomb 'Acme web', resolve them and leave a one-line comment explaining it shipped in v1.5."

The AI will:

1. list_bugs({ status: "open", severity: "critical" }).
2. Filter the result by project_name === "Acme web".
3. For each: comment_on_bug({ bug_id, body: "Shipped in v1.5." }) then resolve_bug({ bug_id }).

Mark spam from a known sender

"Mark every bichito whose reporter email is noreply@example.com as spam."

1. list_bugs({}) (or with search if you have a hint).
2. Filter by reporter_email.
3. mark_spam({ bug_id }) per match.

Assign yourself the unassigned bugs you commented on this week

"Find unassigned bichitos I've commented on in the last 7 days and assign them to me."

The AI uses list_bugs({ assignee_id: "none" }) then walks each detail with get_bug to check comments and dates, then assign_me for matches.

Labels

Tag every regression mention

"Tag bichitos whose title or description contains 'regression' with the regression label, create the label if it doesn't exist yet."

1. list_labels({ team_id }) → check whether regression exists.
2. If not, create_label({ team_id, name: "regression", color: "#ef4444" }).
3. list_bugs({ search: "regression" }).
4. attach_label({ bug_id, label_name: "regression" }) per result.

Rename a label across the catalogue

"Rename the label 'urg' to 'urgent' in this hive."

1. list_labels({ team_id }) → grab the id.
2. update_label({ label_id, name: "urgent" }).

Wipe a label

"Delete the 'old-label' label — I'm not using it anymore."

delete_label({ label_id }). The label is removed from every bichito it was attached to in the same call.

Setting up

Install bichito in this Next.js project

"Install bichito in this project. The widget should appear in the bottom-right corner."

1. detect_framework({ repo_root: "<absolute path>" }).
2. verify_api_key() — confirms BICHITO_API_KEY works.
3. get_install_snippet({ framework: "next" }).
4. AI edits the file (app/layout.tsx for App Router, pages/_app.tsx for Pages Router) and pastes the snippet.

Create a new honeycomb on the fly

"Create a honeycomb called 'Acme mobile' for the iOS app."

1. create_honeycomb({ team_id, name: "Acme mobile", description: "iOS app" }) — returns the new project including its sk_… key.
2. The AI can immediately reference the key for the next step.

Reporting

Daily standup line

"Give me a one-line summary of how my hive is doing today: open bichitos, criticals, anything new since yesterday."

get_stats() plus a small list_bugs({ ... }) for the "since yesterday" filter (the AI filters client-side on created_at).

Find the noisy honeycomb

"Which of my honeycombs has the most open bichitos right now?"

get_stats() returns counts grouped by team / honeycomb — the AI just sorts.

Tips

• Be specific about the date range. AI assistants are good at filtering once they have the data; they're worse at guessing what "lately" means.
• Confirm before destructive actions. Most tools are idempotent, but delete_label and bulk resolve aren't reversible. A "first show me what would be affected, then ask for confirmation" prompt is good practice.
• Use the rich docs page once. /docs/mcp has a Copy-as-Prompt button that hands the AI a full briefing of the install + tool surface — save yourself the typing the first time you set things up.