MCP Builder

Best for

• Planning a narrow MCP tool surface for an agent workflow
• Implementing or upgrading MCP servers
• Defining tool/resource contracts before writing code
• Auditing MCP server ergonomics, validation, and failure modes

Workflow

1. Start from the user outcome or agent workflow.
2. Define the minimum viable tool surface before implementation.
3. For each tool, specify:
   • name
   • description
   • input schema
   • return shape
   • failure cases
4. Decide what belongs in:
   • tools
   • resources
   • prompts
5. Choose transport last, after the contract is clear.
6. Validate with at least one realistic usage path.

Output contract

• a tool plan
• implementation approach
• security and auth notes
• test strategy
• rollout or migration notes if applicable

Critical rules

1. Descriptive tool names —

   search_users

   not

   query1

   ; agents pick tools by name
2. Schema first — Every tool input validated, optional params have defaults
3. Structured output — Return machine-usable data and readable errors
4. Fail gracefully — Never leak raw stack traces to clients
5. Stateless by default — Do not rely on call order unless explicitly designed that way
6. Small v1 — Prefer 3-5 sharp tools over a broad vague surface

Starter prompts

• Plan an MCP server for this workflow. Start with the smallest useful tool surface and specify tool contracts before implementation.
• Audit this MCP server for poor tool names, weak schemas, auth gaps, and confusing outputs.
• Implement an MCP server for

  <use case>

  with typed inputs, clear failures, and a realistic test path.

Autonomous decision rules

• the task mentions MCP servers, tools, resources, prompts, or agent connectors
• the user needs agent capabilities exposed through a protocol boundary

• the task is only general API design with no MCP surface
• the user only needs product planning and no protocol/tool design