mcp-wiring

MCP Wiring

Configure MCP (Model Context Protocol) server connections for a plugin by scanning skill requirements, matching to available servers, and generating a validated .mcp.json file.

Workflow

1. Identify plugin — get the plugin directory path
2. Scan requirements — detect what MCP tools the skills need
3. Match servers — map requirements to available MCP server packages
4. Generate config — produce .mcp.json with correct entries
5. Validate — verify the generated config is well-formed
6. Review with user — confirm server choices and env var requirements

Step 1: Identify Plugin

Accept the plugin directory path. The .mcp.json file will be created/updated at the plugin root, alongside .claude-plugin/plugin.json.

If the plugin already has a .mcp.json, read it first — merge new entries rather than overwriting.

Step 2: Scan Requirements

Run the MCP requirements scanner:

python ${CLAUDE_SKILL_DIR}/../../scripts/scan_mcp_requirements.py <plugin-dir>

This scans all SKILL.md files to detect:

• allowed-tools frontmatter entries referencing MCP tools (e.g., mcp__server__tool)
• Tool references in body text suggesting external capabilities
• Capability keywords (database, web search, Slack, etc.)

Capture the JSON output — it contains matched servers, unmatched references, and capability hints.

Step 3: Match Servers

Map each requirement to a known MCP server:

Requirement	MCP Server	Package
Filesystem access	@anthropic/mcp-filesystem	npx -y @anthropic/mcp-filesystem <path>
Web search	@anthropic/mcp-web-search	npx -y @anthropic/mcp-web-search
GitHub access	@modelcontextprotocol/server-github	npx -y @modelcontextprotocol/server-github
Database (PostgreSQL)	@modelcontextprotocol/server-postgres	npx -y @modelcontextprotocol/server-postgres
Slack	@modelcontextprotocol/server-slack	npx -y @modelcontextprotocol/server-slack

For the latest catalog, search the MCP registry if available:

Use tool: mcp__mcp-registry__search_mcp_registry

If a requirement has no known server match, flag it for the user.

Step 4: Generate Config

Pipe the scan output into the config generator:

python ${CLAUDE_SKILL_DIR}/../../scripts/scan_mcp_requirements.py <plugin-dir> | \
python ${CLAUDE_SKILL_DIR}/../../scripts/generate_mcp_json.py <plugin-dir> --merge

The --merge flag preserves existing entries in .mcp.json. The script generates the standard structure:

{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["-y", "@package/name", "..."],
      "env": {
        "KEY": "${ENV_VAR}"
      }
    }
  }
}

Use ${CLAUDE_PLUGIN_ROOT} for plugin-relative paths. Use ${CLAUDE_PLUGIN_DATA} for persistent data paths.

Step 5: Validate

Run the MCP config validator:

python ${CLAUDE_SKILL_DIR}/../../scripts/validate_mcp_config.py <plugin-dir>

This checks: .mcp.json is valid JSON, each server has a command field, environment variables are documented, and the plugin manifest references .mcp.json.

Step 6: Review with User

Present the generated config:

1. List each MCP server and what it provides
2. Highlight environment variables that need to be set
3. Note any unmatched requirements
4. Confirm the user wants to write the config

After approval, update the plugin manifest to reference .mcp.json if not already declared.