skills.rest
Deepak-ai-93
Deepak-ai-93COMMUNITY
Explore all skills by this author →
GitHub

mcp-agent-builder

Scaffolds runnable MCP servers and agent wiring from a discovery interview through audited delivery.

1|Updated Aug 7, 2026
Search Tags:#workflow automation#scaffolding#claude code#api integration#model context protocol#mcp server#agent builder
One-click install
npx skills add https://github.com/Deepak-ai-93/deepak-skill --skill mcp-agent-builder-deepak-ai-93
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mcp-agent-builder
Source: https://github.com/Deepak-ai-93/deepak-skill/tree/main/skills/mcp-agent-builder
Command: npx skills add https://github.com/Deepak-ai-93/deepak-skill --skill mcp-agent-builder-deepak-ai-93

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires @modelcontextprotocol/sdk, zod, and includes scripts (resource) components.

What problem does it solve? Building an MCP server from scratch involves many decisions—deployment model, transport, auth flow, tool design, and client configuration—that are easy to get wrong, producing servers nobody can run or connect. This Skill turns an "I want an agent that can do X" request into a complete, audited build pack covering everything from requirements to a runnable server. ## Core Features & Use Cases - Structured 7-stage workflow: discovery interview → mcp-prd.md → mcp-architecture.md → mcp-plan.json → scaffolded server → agent-design.md + ide-cli-matrix.md → automated audit plus a fresh auditor subagent scoring the pack out of 50. - Real server generation: scaffold-server.mjs validates the plan (kebab-case name, deployment model, auth mode, tool schemas) and generates a working @modelcontextprotocol/sdk server with zod-typed handlers, README, per-client config snippets, and .env.example. - Connection matrix for 9+ clients: exact connect commands for Claude Code, Cursor, Codex CLI, Gemini CLI, OpenCode, Cline, Windsurf, VS Code, and Zed, covering both stdio and remote HTTP deployments. - Use Case: Ask your agent to "build an MCP server that posts approved drafts to Instagram and X" and receive a full pack: PRD, architecture with OAuth 2.0 flow, a runnable server with post_draft/get_analytics/list_drafts tools, agent system prompt, and connect commands. ## Quick Start Ask your AI agent to build an MCP server for the capability you need, and it will run the discovery interview before scaffolding and auditing the deliverable pack.

Frequently Asked Questions about mcp-agent-builder

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build an MCP server for Claude Code or Cursor?▼

Run the discovery interview to lock the deployment model, auth, and tool surface, then write mcp-plan.json and run scaffold-server.mjs. It generates a runnable @modelcontextprotocol/sdk server with connect commands like claude mcp add for each client.

What is the difference between stdio and remote HTTP MCP deployment?▼

Stdio runs locally for IDE/CLI use and is started with npm start, while remote HTTP exposes one URL for everyone and supports OAuth 2.0 redirect flows. The architecture template requires picking exactly one model, with transport and auth matching it.

Can I use OAuth 2.0 with a stdio MCP server?▼

OAuth 2.0 redirect flows do not work over stdio, which the skill calls the OAuth-on-stdio trap. For OAuth, choose remote HTTP deployment, or document a device-code or CLI flow explicitly in the architecture.

How do I connect an MCP server to Codex CLI or Gemini CLI?▼

The generated ide-cli-matrix.md provides exact commands per client: a [mcp_servers] block in ~/.codex/config.toml for Codex CLI, and gemini mcp add for Gemini CLI. The scaffold also writes ready-to-copy config snippets under server/config/.

Why does scaffold-server.mjs exit with an error on my plan?▼

The script validates mcp-plan.json and exits 1 on failures such as a non-kebab-case name, an unknown deployment or auth mode, zero tools, tool descriptions under 20 characters, or param types outside string, number, and boolean. Fix the listed issues and re-run.

What does the MCP pack audit check before delivery?▼

audit-mcp.mjs verifies PRD and architecture sections, SDK dependency, one handler per plan tool, connect steps, .env.example, and absence of committed secrets. A fresh auditor subagent then scores the pack out of 50, with 35 or higher required to ship.

Related Skills

Explore More Software Engineering Skills→
hasanraiyanhasanraiyan
12

build-mcp-server

Guide MCP server architecture and implementation handoff for Claude compatibility.

Community
Advanced
ecoDriverltdecoDriverltd

mcp-csharp-publish

Publish C# MCP servers to NuGet, Docker, Azure, and the MCP Registry.

Official
Advanced
ldmrepoldmrepo
3

agui

Build real-time streaming AI agent applications using the AG-UI protocol.

Community
Intermediate
Rawgrowth-ConsultingRawgrowth-Consulting

fastmcp

Create, validate, and deploy FastMCP MCP servers in Python.

Official
Advanced