MCP — overview

The bichito MCP server is a small npm package (@bichito/mcp) that exposes bichito's read/write surface to any Model Context Protocol client — Claude Code, Cursor, Continue, and others. Once installed, your AI assistant can:

• Install the widget in any codebase — detect the framework, verify your API key, paste the right snippet in the right file.
• Read your inbox — list / filter bichitos, get details, get cross-team stats.
• Triage — resolve, mark spam, assign to yourself, comment.
• Manage labels — create, rename, delete, attach to bichitos, detach.
• Update bichitos — set status and severity from chat.

Every mutation made over MCP records via: "mcp" in the activity timeline so you can tell AI-driven changes apart from human ones.

Why is this useful?

A few examples that take seconds with MCP and minutes without:

• "Show me my critical bichitos from this week and resolve the ones from honeycomb X."
• "Tag every bichito whose title contains 'login' with the regression label, create the label if it doesn't exist."
• "Mark all reports from noreply@example.com as spam."
• "Install bichito in this Next.js project."

What's in the rest of this section

• Install — Claude Code, Cursor, Continue.
• Scopes — what each scope unlocks; the /settings/mcp UI walkthrough.
• Tools — every tool, its inputs, its required scope, error cases.
• Recipes — common AI-driven flows you can copy.

There's also a rich install page at /docs/mcp (the React-rendered one with a Copy-as-Prompt button so your AI gets full context in one paste). Use that one when you want to bootstrap the AI quickly.