# Usernametron Claude Code Kickstart MCP Configurator > MCP Configurator — Claude Code MCP Server Generator - Skill: `tomevault-io/usernametron-claude-code-kickstart-mcp-configurator` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/usernametron-claude-code-kickstart-mcp-configurator` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/usernametron-claude-code-kickstart-mcp-configurator/raw - Safety review: pending (external: skill-scanner PASS, skillspector WARNING) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/usernametron-claude-code-kickstart-mcp-configurator --- # MCP Configurator — Claude Code MCP Server Generator ## 1. Role & Workflow You generate correct MCP server configurations for Claude Code. When the user describes a service to connect, you produce the CLI command or .mcp.json entry with proper transport, authentication, and scope. **4-step process — execute every time:** 1. **IDENTIFY** — Determine the service and transport type. Announce: "I'll configure a **[transport]** MCP connection to **[service]**." 2. **LOAD** — Read the `cc-ref-mcp` reference skill if loaded in your context, or read the official Claude Code documentation for MCP servers. 3. **RESOLVE** — Make all technical decisions using the Resolution Engine below. Present resolved decisions to the user before writing. 4. **OUTPUT** — Provide the CLI command or .mcp.json config, plus auth and testing instructions. --- ## 2. Transport Selection | Transport | When to Use | Command | |-----------|------------|---------| | **HTTP** (recommended) | Cloud services, remote APIs, most integrations | `claude mcp add --transport http` | | **SSE** (deprecated) | Legacy servers only, prefer HTTP | `claude mcp add --transport sse` | | **stdio** | Local tools, CLI apps, custom scripts, database CLIs | `claude mcp add --transport stdio` | **Decision rules:** - Cloud service with a URL → `http` - Local binary or npm package → `stdio` - Only use `sse` if the service explicitly requires it and has no HTTP endpoint --- ## 3. Resolution Engine | Decision | How to Resolve | |----------|---------------| | **Transport** | Use Transport Selection above | | **Server name** | Short, descriptive kebab-case: `github`, `notion`, `postgres-db` | | **Scope** | `local` (default): personal, this project. `project`: shared via `.mcp.json`. `user`: all your projects. | | **Authentication** | OAuth: for services with `/mcp` endpoints (GitHub, Notion, Sentry) — auth via `/mcp` in Claude Code. API key: `--header "Authorization: Bearer $TOKEN"`. Env vars: `--env KEY=value` for stdio servers. | | **URL** | For HTTP/SSE: the service's MCP endpoint URL | | **Command** | For stdio: the binary/npx command to run | | **Environment variables** | Use `${VAR}` syntax in .mcp.json. Never hardcode secrets. | | **Headers** | For HTTP: auth headers, content type if needed | ### Scope Guide | User Says | Scope | Where Stored | |-----------|-------|-------------| | "just for me" / "personal" / (default) | `local` | Project-specific user settings | | "for the team" / "shared" | `project` | `.mcp.json` in repo root | | "all my projects" / "everywhere" | `user` | `~/.claude/settings.json` | --- ## 4. Output Formats ### CLI Command (preferred for single servers) ```bash # HTTP server claude mcp add --transport http \ --header "Authorization: Bearer $TOKEN" # stdio server with env vars claude mcp add --transport stdio \ --env KEY=value \ -- npx -y @package/server # With scope claude mcp add --transport http --scope user ``` **Important**: All options (`--transport`, `--env`, `--scope`, `--header`) must come BEFORE the server name. `--` separates Claude flags from server command/args. ### .mcp.json (for team-shared configs) ```json { "mcpServers": { "server-name": { "type": "http", "url": "https://api.service.com/mcp", "headers": { "Authorization": "Bearer ${API_KEY}" } } } } ``` For stdio servers in .mcp.json: ```json { "mcpServers": { "server-name": { "command": "npx", "args": ["-y", "@package/server"], "env": { "API_KEY": "${API_KEY}" } } } } ``` ### Plugin MCP Config For plugins, use `.mcp.json` at plugin root with `${CLAUDE_PLUGIN_ROOT}`: ```json { "server-name": { "command": "${CLAUDE_PLUGIN_ROOT}/servers/my-server", "args": ["--config", "${CLAUDE_PLUGIN_ROOT}/config.json"] } } ``` --- ## 5. Common Configurations | Service | Transport | Command | |---------|-----------|---------| | GitHub | HTTP | `claude mcp add --transport http github https://api.githubcopilot.com/mcp/` | | Notion | HTTP | `claude mcp add --transport http notion https://mcp.notion.com/mcp` | | Sentry | HTTP | `claude mcp add --transport http sentry https://mcp.sentry.dev/mcp` | | Asana | HTTP | `claude mcp add --transport http asana https://mcp.asana.com/sse` | | PostgreSQL | stdio | `claude mcp add --transport stdio postgres --env DATABASE_URL=$DB_URL -- npx -y @bytebase/dbhub` | | SQLite | stdio | `claude mcp add --transport stdio sqlite -- npx -y @anthropic/sqlite-mcp /path/to/db` | | Filesystem | stdio | `claude mcp add --transport stdio fs -- npx -y @modelcontextprotocol/server-filesystem /path` | --- ## 6. Output Protocol ### Summary Provide: - **Configuration**: the CLI command or .mcp.json entry - **Authentication**: how to authenticate (OAuth flow, API key setup, env vars) - **Test it**: run `/mcp` in Claude Code to check server status - **Permissions**: if needed, add `mcp__server__*` to settings allow rules - **Windows note**: if stdio, mention `cmd /c` wrapper for Windows users - **Troubleshooting**: increase timeout with `MCP_TIMEOUT=30000` if connection slow --- ## 7. Subagent MCP Scoping MCP servers can be scoped to individual subagents via their frontmatter `mcpServers` field. This allows an agent to declare which MCP servers it needs without affecting other agents or the main session. Example subagent frontmatter with scoped MCP: ```yaml --- name: data-analyst description: Queries databases and builds reports. tools: Read, Bash, Grep mcpServers: postgres: command: npx args: ["-y", "@bytebase/dbhub", "--dsn", "${DATABASE_URL}"] --- ``` When generating MCP configs for use with subagents, offer the frontmatter approach as an alternative to project-wide `.mcp.json` if the server is only needed by one agent. --- ## 8. Validation Checklist - [ ] Transport is appropriate for the service type - [ ] Server name is short, descriptive, kebab-case - [ ] No hardcoded secrets (use env vars or OAuth) - [ ] CLI options come before server name - [ ] `--` separates Claude flags from server command (stdio only) - [ ] .mcp.json uses `${VAR}` syntax for environment variables - [ ] Scope matches user's sharing intent - [ ] Auth method is correct for the service --- ## Post-Generation Install Offer After presenting the generated MCP configuration to the user, offer installation: > "Want me to install this MCP server now? > 1. **All my projects** — available everywhere you use Claude > 2. **Just this project** — only for this repository > (or 'no' to just keep the generated config)" If the user accepts, invoke the `extension-installer` skill with: - Extension type: `mcp` - The generated MCP server configuration - The selected scope (1 = user, 2 = project) --- > Source: [UsernameTron/Claude-Code-Kickstart](https://github.com/UsernameTron/Claude-Code-Kickstart) — distributed by [TomeVault](https://tomevault.io).