MCP — install

The bichito MCP server runs locally on the user's machine and speaks MCP over stdio. It needs two pieces of credentials, both available in the dashboard:

• BICHITO_MCP_TOKEN — your per-user MCP token (mcp_…). Generate one in /settings/mcp. Pick the scopes you want it to carry (see Scopes).
• BICHITO_API_KEY — your honeycomb's project API key (sk_…). Only needed for the install-the-widget tools (get_install_snippet, verify_api_key). Skip if you only want triage tools.

If you maintain multiple hives (e.g. one per client), you can register the MCP server multiple times in the same AI tool with different env-var sets — each pointing at a hive-scoped token (see Scopes). One Claude Code instance can carry several bichito connections, one per hive, with no cross-bleed.

Claude Code

claude mcp add bichito \
  --env BICHITO_API_KEY=sk_xxx \
  --env BICHITO_MCP_TOKEN=mcp_xxx \
  -- npx -y @bichito/mcp

Drop either env var if you only need one half of the tools. Restart Claude Code after the command — the new MCP server only registers on relaunch.

Cursor / Continue / others

Edit your tool's MCP config file (typically ~/.config/<tool>/mcp.json or under your IDE's settings) and add:

{
  "mcpServers": {
    "bichito": {
      "command": "npx",
      "args": ["-y", "@bichito/mcp"],
      "env": {
        "BICHITO_API_KEY": "sk_xxx",
        "BICHITO_MCP_TOKEN": "mcp_xxx"
      }
    }
  }
}

Restart the editor. The bichito tools should appear in the available-tools list immediately.

Pinning a version

npx -y @bichito/mcp always installs the latest published version. If you want determinism (e.g. CI, shared team config), pin:

npx -y @bichito/mcp@0.1.5

Troubleshooting

"401 Invalid MCP token" — the token you pasted is wrong or revoked. Mint a new one at /settings/mcp and update BICHITO_MCP_TOKEN.

"403 scope required: write:bugs" — your token doesn't carry the scope this tool needs. Re-mint with the scope checked, or use a tool that doesn't need write access.

The new tools don't appear after npm publish — restart your AI tool. The MCP server is launched once per session; new tool definitions load on the next launch. If npx is caching an old version, force the resolution: rm -rf ~/.npm/_npx or pin a specific version.