MCP Builder · Connector studioA-01 InformationA-02 InstructionA-03 BenchmarkA-04 Showcase

A-01 General arrangement

Make an APIspeak agent.

Build MCP servers around the work an assistant actually needs to do. Anthropic's MCP Builder takes your coding assistant through research, implementation, review and evaluation in one practical development guide.

Fig. 18-second capability overview. Illustrative animation; test evidence appears separately.
01The interface

One tool, exploded.

An MCP server sits between your service and an assistant. The guide treats each tool as an assembly of deliberate parts, and the quality of the server depends on how well an assistant can find, call and recover with those parts.

Fig. 2 · Tool anatomyParts per bundled best-practice & SDK guides
Your serviceEndpoints · authentication · data modelsSocket
  1. Namegithub_create_issue
    A

    Discoverable nameService prefix, action first: {service}_{action}_{resource}, so tools stay distinct alongside other servers.

  2. Input schemaZod (TypeScript) · Pydantic (Python)
    B

    Typed inputConstraints, clear descriptions and examples inside the field descriptions.

  3. OutputoutputSchema · structuredContent
    C

    Structured resultsStructured data alongside text, with JSON or Markdown response formats.

  4. AnnotationsreadOnlyHint · destructiveHint · idempotentHint · openWorldHint
    D

    Declared behaviourHints that tell clients what a call can change.

  5. Context & errorslimit · has_more · next_offset
    E

    Focused, recoverablePaginated, relevant data and error messages that suggest the next step.

AssistantDiscovers · calls · recoversPort
Fig. 2Diagram of the guidance, not generated output. Example name taken from the bundled best-practice guide.
02Fit

For developers turning APIs into tools.

It suits internal or external services that an assistant should be able to query and act on, built in TypeScript or Python with a coding assistant doing the work.

Built for

  • Wrapping an external API or service as an MCP server, in TypeScript (MCP SDK) or Python (FastMCP).
  • New TypeScript servers on SDK v2, or existing projects that already depend on SDK v1.
  • Choosing between stdio for local servers and Streamable HTTP for remote ones.
  • Writing a read-only evaluation set that tests whether a model can use the finished server.

Not built for

  • A ready-made integration for any particular service.
  • Hosted server infrastructure, deployment or API credentials.
  • A guarantee that generated code is production ready.
  • Use without an assistant that can write files and run your toolchain.
03Bill of materials

Bring the service. Define the outcome.

No.PartPurpose
Supplied by you
aAPI documentation and target workflowsWhat the tools must let an assistant accomplish.
bAn authorized authentication methodHow the server is permitted to call your service.
cA Python or TypeScript projectIncluding any existing SDK version, deployment and transport constraints.
In the package 11 original files
1SKILL.mdThe four-phase development guide.
2reference/mcp_best_practices.mdNaming, response formats, pagination, transport, security, annotations, errors.
3reference/node_mcp_server_v2.mdTypeScript SDK v2 guide with a self-check script for the built server.
4reference/node_mcp_server.mdThe equivalent guide for existing SDK v1 projects.
5reference/python_mcp_server.mdPython / FastMCP patterns and a quality checklist.
6reference/evaluation.mdQuestion design, answer verification and runner usage.
7scripts/evaluation.py · connections.pyPython evaluation client for stdio, SSE and HTTP servers.
8scripts/requirements.txtEvaluation client dependencies.
9scripts/example_evaluation.xmlFive-question format example.
10LICENSE.txtApache License 2.0.
Packaging additions
+PROVENANCE.md · UPSTREAM-THIRD-PARTY-NOTICES.mdSource, pinned commit, attribution and upstream-wide notices.
+RUNTIME-NOTES.md · evaluation-constraints.txtThe tested Python dependency constraint for the evaluator.
Produced by your assistant not shipped
→Server implementationAPI client, error helpers, response formatting and pagination.
→Tool contractsTyped schemas, useful descriptions and annotations.
→Build and inspection checksCompilation or syntax checks and MCP Inspector testing.
→Evaluation XMLTen verified, read-only questions, as the guide specifies.
04Requirements

Dependencies & boundaries.

No Skill-specific API key, but not free of model use.

A coding assistant that can read Skill folders, write files and run commandsAlwaysUses your host assistant's own model access and plan.
A Node/TypeScript or Python toolchainBuild & testDependency installs need network access.
Network access to official MCP and SDK docsResearch & buildThe guide fetches current specifications and SDK READMEs.
Credentials for the service you integrateImplement & testYour responsibility, under that service's own terms.
Your own Anthropic API access and a target MCP serverLive evaluation onlyModel requests are billed to your account.
One compatibility detail mattersThe unchanged Python evaluator fails with MCP Python 2.3.0. The package includes an installation constraint tested with MCP Python 1.30.0. It does not restrict the separate TypeScript v2 server guide.

Keep secrets out of prompts and source archives. Evaluation questions should use stable, read-only tasks. Review server behaviour and access scopes before you connect anything to real systems.

05Verified

What has been checked.

11upstream files copied byte for byte
4offline helper checks passed in a local smoke run
0live model-backed evaluations run

Python syntax, evaluator help, example XML parsing and response extraction passed. Full server generation and a model-backed evaluation have not been tested.

The full inspection record and run evidence are on sheet A-03, Benchmark.