VividKit Crystal clear AI coding
VI
Choose Your Setup Get Started with AgentKit Continue with ClaudeKit Migrate from ClaudeKit to AgentKit Clean cutover from ClaudeKit What Is AgentKit? How AgentKit Works AgentKit Kits, Skills, Agents & Hooks AgentKit Runtime Support AgentKit Desktop App AgentKit Configuration AgentKit CLI Quick Reference AgentKit Skills Cheatsheet AgentKit Workflows AgentKit Statusline Update AgentKit CLI & Kits AgentKit Helper Clean reinstall of AgentKit AgentKit Troubleshooting What is ClaudeKit? ClaudeKit CLI Guide Command Finder Commands Cheat Sheet Workflow Recipes CLI Quick Reference Claude Mechanics Permission Modes ClaudeKit Hooks Coexistence ClaudeKit x Codex CK → Other coding tools How ClaudeKit Works Inside ClaudeKit Finding Your Unknowns Session Recovery Fix from Logs UI/UX Guide Accounts Switcher IDE Config Remote Control Codex App System Diagramming UI Review Gate Promotions Support VividKit

Quick Ref / Command

Engineer Kit

/ak:mcp-builder

High-signal MCP servers

Build, upgrade, test, and publish MCP servers that expose external services as tools, targeting the 2026-07-28 protocol with the TypeScript SDK v2 or Python mcp v2 (MCPServer).

01

Start from the task

02

Inspect the stack

03

Pick the transport

04

Build one complete tool

Rule 01

Define the task and first tool contract before infrastructure

Rule 02

Annotations are hints, not enforcement; authorize every operation server-side

Rule 03

One complete, verified tool beats many registered ones

Rule 04

Registration alone is not completion; report what a real client observed

Execution Map

Core lanes from input to output, aligned with the command's real execution path.

01

Start

Input, scope, route

  1. 1 Start from the task Define the user task and the first tool contract (inputs, output shape, side effects, auth) before choosing infrastructure.
  2. 2 Inspect the stack Check the project language, SDK version, and transport. Upgrade a v1 SDK only when asked or when a required feature needs it; read the protocol reference for deprecated features (sampling, roots, logging, SSE).
02

Work

Agent / skill execution

  1. 3 Pick the transport Default to stdio for local single-user servers and Streamable HTTP for remote or multi-user servers. Never start a new server on the deprecated HTTP+SSE transport.
  2. 4 Build one complete tool Build one working path from validated input through authorized upstream calls to bounded, structured output: constrained input schema, outputSchema with structuredContent, accurate annotations, actionable isError results, timeouts, and pagination.
03

Verify

Gate, review, validation

  1. 5 Authorize and confirm Authorize every operation server-side. Confirm destructive actions through elicitation; for clients that cannot prompt, refuse until an explicit confirmation argument arrives. Keep credentials in environment variables or validated OAuth tokens and never pass client tokens upstream.
  2. 6 Verify with a real client Type- or import-check, then drive the server with MCP Inspector CLI or an SDK client: valid calls, invalid input, upstream failures, pagination, and for HTTP 401/403/Origin. Use bounded, harness-owned processes and stop those you started.
04

Close

Report, handoff, artifact

  1. 7 Evaluate task quality Write read-only evaluation questions and run scripts/evaluation.py; add tools only as the requested workflow needs them and rerun checks after each change.
  2. 8 Report, publish on request Report working tools, observed client results, and missing or unverified capabilities; registration alone is not completion. Publish server.json to the MCP Registry only when the user asks.

Syntax, arguments, subcommands

How to invoke the skill: syntax, positional arguments, shared options, then each subcommand with its own syntax and outcome.

Syntax

/ak:mcp-builder [service or API to integrate]

Arguments

[service or API to integrate] Required

Target integration

External service or API to turn into a new MCP server or expanded tool/resource surface. Include intended workflows, language, transport, auth model, permissions, rate limits, data volume, and allowed external effects when known.

/ak:mcp-builder "Build a TypeScript MCP server for the Acme tickets API with read-only search first, cursor pagination, and stdio transport"

Sample Prompt

Concrete invocations for each flag, subcommand, mode, or route available in the live workflow.

Default API server Recommended 
/ak:mcp-builder Linear issue triage API
Use when:
Invoke when building an MCP server or tool surface for an external service.
Expected:
Defines the triage task and first tool contract, builds one complete tool with structured output and annotations, drives it with MCP Inspector, then adds tools the workflow needs.
Python MCPServer 
/ak:mcp-builder Python mcp v2 server for customer search over stdio
Use when:
Use when the server should be built in Python with the mcp v2 SDK.
Expected:
Uses the Python server reference with MCPServer, defines constrained inputs and an outputSchema, keeps credentials in environment variables, and verifies valid, invalid, and upstream-failure calls.
Remote TypeScript server 
/ak:mcp-builder TypeScript server for billing admin workflows over Streamable HTTP with OAuth
Use when:
Use when the host project needs a multi-user remote server on @modelcontextprotocol/server v2.
Expected:
Builds on Streamable HTTP, validates OAuth tokens for this server, confirms destructive billing actions through elicitation, and tests 401, 403, and Origin handling.
Upgrade a v1 server 
/ak:mcp-builder migrate our v1 SSE server to the 2026-07-28 protocol
Use when:
Use when an existing server relies on v1 SDKs or deprecated features such as SSE, sampling, roots, or logging.
Expected:
Reads the protocol migration reference, moves the transport off HTTP+SSE, upgrades the SDK, and reruns Inspector checks and evaluations to show the tools still behave the same.

Handled Scope

  • MCP spec 2026-07-28
  • TypeScript SDK v2 and Python mcp v2 servers
  • Tool design, schemas, and annotations
  • HTTP auth and OAuth
  • Inspector checks and evaluations
  • MCP Registry publishing

Artifacts Produced

MCP Deliverables

Pattern

Server code + tool schemas + client results + evaluations

Working tools, observed client results from Inspector or an SDK client, evaluation outcomes, and any missing or unverified capabilities; a registry entry only when publishing was requested.

Next