mcp-engineering

Build MCP servers with well-named tools, workflow-oriented coverage, and clear errors. Use when exposing capabilities to agents via MCP. Pair with security/agent-security for the audit side.

SKILL.md

GitHub

mcp-engineering

Build the tool the agent wishes existed: workflow-shaped, narrowly permissioned, loudly correct on misuse.

Activate when

  • Creating or extending an MCP server (Python FastMCP or TypeScript).
  • Deciding tool boundaries, naming, or error formats.

Do NOT activate for

  • Auditing an existing MCP server for security (see security/agent-security).

Procedure

  1. Scope: one tool per user-meaningful workflow step, not per internal function; prefer fewer, composable tools over exhaustive coverage.
  2. Naming: verb-object names, typed inputs, documented preconditions; errors state what was wrong, what was expected, and the safe retry.
  3. Permissions: least privilege per tool; destructive tools require confirmation params and log intent.
  4. Versioning: additive changes first; breaking tool-shape changes get a version bump + migration note.
  5. Evaluate: scripted happy-path + misuse + adversarial-input runs before shipping.
  6. Verify: run the eval script, exercise tools from a real client, and confirm logs show who called what.

References

  • Related: ../agent-architecture/, ../../security/agent-security/.

Related in security

git-workflow

Git workflow guidance for commits, branches, and pull requests

by agno-agi
37,144

clerk-observability

Implement monitoring, logging, and observability for Clerk authentication. Use when setting up monitoring, debugging aut...

by jeremylongshore
1,061

juicebox-migration-deep-dive

Advanced Juicebox data migration strategies. Use when migrating from other recruiting platforms, performing bulk data im...

by jeremylongshore
1,061