RetentaRetenta

Evidence record · captured 2026-10-10

Anthropic MCP Builder

A four-phase guide to researching, implementing, testing, and evaluating MCP servers with discoverable tools, typed schemas, focused responses, and realistic read-only evals.

Official publisherUnverifiedSource reviewedBehavior not tested
Source review

Editorial assessment

This is a source-only assessment. It does not claim runtime or security testing.

What it does

The source balances endpoint coverage with higher-level workflows, then walks through API research, transport selection, typed tools, concise responses, build and inspector checks, and a ten-question evaluation set.

Best for

  • Designing a new MCP server around an external API
  • Improving tool names, schemas, errors, and pagination
  • Creating task-level evaluations for an MCP integration

Strengths in the source

  • Treats agent discoverability and context size as first-class API-design concerns.
  • Calls for structured outputs and behavioral annotations, including destructive and read-only hints.
  • Extends implementation through realistic, verifiable evaluations instead of stopping at compilation.

Limitations

  • The workflow depends on current protocol and SDK references beyond the top-level file.
  • Comprehensive endpoint coverage and ten verified evaluations can be expensive for a small integration.
  • The preferred stack is opinionated toward TypeScript and streamable HTTP.

When not to use it

  • A local helper with no need for the MCP protocol
  • An integration whose API cannot support stable, permission-aware tools

Reviewed by Retenta editorial on 2026-10-10; scope: source only.

Files included in this review
  • skills/mcp-builder/SKILL.md
  • README.md
  • skills/mcp-builder/LICENSE.txt

Evidence and boundaries

Execution · not tested

Retenta reviewed the pinned source but did not install or execute this skill.

Security · not tested

No sandbox run, dependency audit, or independent security assessment is claimed.

Pinned publisher source
skills/mcp-builder/SKILL.md
Revision
dbd4588f9e1033efb41dad4bef2f7947c8993d44
License evidence
Apache-2.0The pinned skill directory includes an Apache License, Version 2.0 LICENSE.txt.
Last source change
2025-12-01
Independent verification
The source is controlled by Anthropic; Retenta has not independently verified its runtime behavior or security.
Upstream repository
Open publisher repository

Compatibility

Claude Code

documented

Anthropic documents these skills as Claude implementations distributed through its Claude Code plugin marketplace.

Requirements and access

  • An external service or API to expose
  • Current MCP specification and SDK documentation
  • TypeScript MCP SDK or Python FastMCP tooling

Referenced tools

MCP InspectorTypeScript SDKPython SDKZodPydantic

Install from the source

Anthropic documents this skill inside the example skill set; the command installs that set rather than one folder.

View the SKILL.md at its source Verify the publisher's install documentation

More AI development skills →

Claude Code plugin install

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

Thyme RAG Capsules

Make this skill’s state durable, governed, and portable.

A Retenta capsule lets agents recall prior work, append provenance-stamped progress, and carry the memory to the next run.

Create durable agent memoryUnderstand the evidence labels