| id | mcp-builder |
| name | MCP Builder |
| description | Guides creation of MCP servers with strong tool design, schemas, and evaluation practices. |
| category | core |
| type | core |
| version | 1.0.0 |
| license | MIT |
| metadata | {"author":"FreakingJSON","source":"OBSOLETE/migration"} |
| compatibility | ["OpenCode","Claude","Gemini","Codex"] |
MCP Builder Skill
Build high-quality MCP (Model Context Protocol) servers that expose reliable tools for LLMs.
Instructions
1. Planning
- Balance full API coverage with workflow tools.
- Use descriptive tool names with consistent prefixes (e.g. github_create_issue).
- Keep tool descriptions concise and actionable.
- Design outputs that are focused, paginated, and easy to parse.
- Provide actionable error messages with next steps.
2. Architecture
- Prefer TypeScript with the MCP SDK when possible.
- Use streamable HTTP for remote servers and stdio for local servers.
- Define shared utilities: auth, error handling, pagination, response formatters.
3. Tool Design
- Define input schemas (Zod or Pydantic) with constraints and examples.
- Define output schemas when possible for structured results.
- Include annotations: readOnlyHint, destructiveHint, idempotentHint, openWorldHint.
- Return both text and structured content when supported.
4. Implementation
- Use async I/O and consistent error handling.
- Ensure pagination on list endpoints.
- Keep tools atomic and predictable.
5. Review and Test
- Compile/build the server and run basic checks.
- Test with MCP Inspector or client smoke tests.
- Verify tool discoverability and error paths.
6. Evaluation
- Create 10 realistic, read-only evaluation questions.
- Ensure each question is verifiable and stable.
- Use a simple XML format with qa pairs.
References
Output Format
- Provide a build plan, tool list, schemas, and testing checklist.
Supported Commands
- "Design an MCP server for: "
- "Draft tool schemas for: "