Add MCP Server Command

You are helping the user add an MCP server to their Claude Code and Codex configurations.

Arguments Parsing

Parse $ARGUMENTS for MCP server details:

• Server name: Look for a clear identifier (e.g., "context7", "github", "figma")
• Transport type:
  • STDIO if mentions "npx", "node", "python", a command, or local execution
  • HTTP if mentions "https://", a URL, or remote/cloud
• Command/URL: The actual command (for STDIO) or URL (for HTTP)
• Environment variables: Look for "env", "token", "key", or KEY=VALUE patterns

Flags

Check for these flags in $ARGUMENTS:

• --claude-only: Only configure Claude Code
• --codex-only: Only configure Codex
• --project: Use project scope instead of user scope (default: user scope)

Flow

Step 1: Extract Information

If the user provided sufficient context (server name + command/URL), proceed to Step 2.

If missing required information, ask conversationally:

"I'll help you add an MCP server. I need a few details:

• What should this server be called? (e.g., 'context7', 'github')
• Is this a local server (runs a command) or remote server (connects to URL)?
• What command should start it / What URL should it connect to?
• Any environment variables needed (like API tokens)?"

Step 2: Validate

Before configuring:

1. For STDIO servers: Check if the command exists (e.g., which npx)
2. For HTTP servers: Validate URL format
3. Validate JSON syntax for Claude config
4. Validate TOML syntax for Codex config

Step 3: Determine Target Files

User Scope (default):

• Claude: ~/.claude.json (add to mcpServers object)
• Codex: ~/.codex/config.toml (add [mcp_servers.NAME] section)

Project Scope (--project flag):

• Claude: .mcp.json in current directory
• Codex: Project scope not supported for Codex MCP - warn user

Step 4: Generate Configurations

For Claude (JSON format):

{
  "mcpServers": {
    "SERVER_NAME": {
      "command": "command here",
      "args": ["arg1", "arg2"],
      "env": {
        "KEY": "value"
      }
    }
  }
}

For HTTP servers, use:

{
  "mcpServers": {
    "SERVER_NAME": {
      "type": "http",
      "url": "https://...",
      "headers": {
        "Authorization": "Bearer ${TOKEN_ENV_VAR}"
      }
    }
  }
}

For Codex (TOML format):

[mcp_servers.SERVER_NAME]
command = "command here"
args = ["arg1", "arg2"]

[mcp_servers.SERVER_NAME.env]
KEY = "value"

For HTTP servers:

[mcp_servers.SERVER_NAME]
url = "https://..."
bearer_token_env_var = "TOKEN_ENV_VAR"

Step 5: Apply Changes

1. Read existing config files (create if they don't exist)
2. Merge new server config (don't overwrite existing servers)
3. Write updated configs
4. If one system fails: Ask user "Claude config succeeded but Codex failed. Keep Claude changes? [y/n]"

Step 6: Confirm

Report success with details:

"Added MCP server 'SERVER_NAME' to:

• Claude: ~/.claude.json
• Codex: ~/.codex/config.toml

Restart Claude Code and Codex to load the new server."

Examples

User input: "Add context7 using npx @upstash/context7-mcp"

• Name: context7
• Transport: STDIO
• Command: npx
• Args: ["@upstash/context7-mcp"]

User input: "Add the Figma MCP at https://mcp.figma.com with FIGMA_TOKEN"

• Name: figma
• Transport: HTTP
• URL: https://mcp.figma.com
• Auth: bearer_token_env_var = "FIGMA_TOKEN"

Reference

For complete format specifications, invoke the meta-unify-core skill to access:

• references/claude-formats.md
• references/codex-formats.md