Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    MCP Server Builder: Ship MCP Servers Agents Use Well

    3

    MCP Server Builder makes your coding agent build servers the way the best ones are built.

    Free

    10 uses5.0 (1 review)
    MCP Server Builder: Ship MCP Servers Agents Use Well

    MCP Server Builder: Ship MCP Servers Agents Use Well

    Example session with this skill installed

    Build an MCP server in TypeScript that lets Claude Code work with our Shopify store: search orders, check inventory and issue refunds. Then connect it to Claude Code.

    • Read your context and instructions
    • Compiled the mcp server builder

    Summary

    Built shopify-mcp (TypeScript, stdio) with 5 task-oriented tools instead of wrapping 60+ Admin API endpoints. Auth via SHOPIFY_ADMIN_TOKEN env var (read_orders, read_inventory, write_orders scopes only).

    Tools

    ToolPurposeType
    search_ordersFind orders by customer, email, status or date range; cursor paginationread-only
    get_orderFull order detail incl. line items and fulfillmentread-only
    check_inventoryStock by SKU or product name across locationsread-only
    low_stock_reportProducts below a threshold, sorted by days of coverread-only
    refund_orderRefund full or specific lines; dry_run=true by defaultdestructive

    Example description (what the model sees):

    Refund an order fully or by line items. Use only after get_order confirms the order and amounts. Do NOT use for cancellations (use Shopify admin). Runs as dry_run unless dry_run=false. Returns refund preview or refund id.

    Safety & robustness

    • Token from env, never logged; startup fails with a clear message if missing
    • Retries with backoff on 429 (Shopify rate limit), 10s timeout
    • Outputs capped at 20 rows with continuation cursor (~600 tokens/call)
    • Errors are actionable: "Order #1043 not found. Use search_orders to find the id."

    Verification

    • tsc --noEmit clean, 14 unit tests passing (client mocked)
    • MCP Inspector: all 5 tools called with valid and invalid input
    • Agent eval: 8/8 tasks picked the right tool; refund task ran dry_run first

    Connect

    npm run build
    claude mcp add shopify --env SHOPIFY_ADMIN_TOKEN=shpat_xxx --env SHOPIFY_STORE=mystore -- node $(pwd)/dist/server.js
    

    Your call

    Decide whether refunds above a set amount should be blocked in the server or left to agent confirmation.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    ## MCP servers that agents actually use correctly Most MCP servers are thin 1:1 API wrappers: 80 tools, vague descriptions, giant JSON outputs. The agent picks the wrong tool, burns context and fails. MCP Server Builder makes your coding agent build servers the way the best ones are built. ## What it does - **Scopes first**: system to wrap, auth, local (stdio) vs remote (Streamable HTTP), TypeScript (`@modelcontextprotocol/sdk`) or Python (`FastMCP`). - **Designs the tool surface before coding**: task-oriented tools (5-15, not 80), `verb_noun` names, descriptions that say what / when / when not / returns, strict zod or pydantic schemas, enums, bounded limits. - **Token-cheap outputs**: concise vs detailed modes, cursor pagination, truncation with continuation hints. - **Safe by default**: secrets from env only, input validation (path traversal, SQL, SSRF), destructive tools with explicit ids and dry_run, tool annotations, auth + Origin checks for remote. - **Robust**: shared client with timeouts and backoff, actionable `isError` messages, nothing on stdout for stdio. - **Verified**: typecheck, unit tests, MCP Inspector on every tool, and a 5-10 task agent eval to tune descriptions. - **Connected**: ready configs for Claude Code (`claude mcp add`), Claude Desktop, Cursor and VS Code, plus a README. - **Debug mode**: step-by-step checklist when your client can't see or misuses a server. ## Includes - SKILL.md workflow - references/tool-design.md - bad vs good tools, description template, output budget, annotations - references/templates.md - TS and Python server templates + client configs - references/checklist.md - pre-delivery checklist and agent eval tasks Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI and any SKILL.md-compatible agent.

    Requires MCP SDKs for TypeScript or Python. Testing requires npx for the MCP Inspector. Transport limited to stdio or Streamable HTTP.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    5.0
    1 review
    5
    1
    4
    0
    3
    0
    2
    0
    1
    0

    This skill has been used 10 times.

    Trust & safety

    Security scan

    Security scan passed 10 days ago

    • Free with an Agensi account

    Requires

    Python
    @modelcontextprotocol/inspector
    Node.js

    Permissions

    Network Access
    Environment Variables
    Listed10 days ago
    Updated5 days ago

    What's inside

    Frequently Asked Questions

    This skill solves common issues like "stdout" log collisions that crash servers, token exhaustion from large payloads, and agent hallucinations caused by vague tool descriptions. It enforces strict validation and pagination patterns that generic prompts typically overlook.

    The skill natively supports TypeScript using the official SDK and Python via FastMCP. It is optimized for use within the Claude Desktop and Cursor environments, ensuring your tools appear correctly in the UI.

    You receive a comprehensive framework for architecting, coding, and testing servers, including optimized templates for tool schemas, error handling logic, and specific instructions for validating your build via the MCP Inspector.

    Yes, the skill includes a specific debugging workflow designed to identify why tools fail to appear in Cursor or Claude, as well as strategies for fixing silent errors in existing stdio or HTTP transport layers.

    While it handles complex API wrapping and resource management, it is currently limited to stdio and Streamable HTTP transport methods; it does not support custom proprietary transport protocols.

    New in LLM Ops