MCP Config Validator (Free)

Validate MCP server configuration files before you restart your client and wonder why a server didn't load, and convert a config from one client's format to another's.

python3 mcp_config_check.py check .mcp.json .cursor/mcp.json .vscode/mcp.json
python3 mcp_config_check.py check ~/Library/Application\ Support/Claude/claude_desktop_config.json --which
python3 mcp_config_check.py check mcp.json --client cursor --json
python3 mcp_config_check.py convert .cursor/mcp.json --to claude-code > .mcp.json

The client is detected from the file name (--client to override). --which also checks that each command exists on your PATH. Exit 1 on errors. Python 3.9+, standard library only, offline. ./demo.sh runs it on the examples; ./run_tests.sh runs 13 test groups.

• Claude Code: a url with no type (Claude Code reads it as stdio and skips it), unknown types (streamable-http is accepted as http), type: "sdk", ${env:X} (Cursor/VS Code syntax) instead of ${X} / ${X:-default}, ${CLAUDE_PROJECT_DIR} without a default in command/args
• Claude Desktop: remote url servers (the JSON file only starts local servers; use Settings > Connectors or a local bridge), relative script paths
• Cursor: envFile on a remote server (stdio only), ${VAR} instead of ${env:VAR}
• VS Code: mcpServers instead of servers, ${input:id} with no matching inputs entry, missing type
• All: invalid JSON with line/column, comments and trailing commas, a whole command line in command ("npx server ."), npx without -y (can hang on the install prompt), @latest packages, non-array args, non-string env, plain http:// to remote hosts, /sse URLs marked as http, hard-coded tokens in files that get committed, the same server in several files, commands not on PATH

convert rewrites the top-level key, adds or removes type, and translates variables: ${VAR} <-> ${env:VAR}, ${CLAUDE_PROJECT_DIR:-.} <-> ${workspaceFolder}. Remote servers are skipped for Claude Desktop with a note; envFile is dropped where it isn't supported, with a note. Review the output before using it.

Checks the documented formats as of November 2026 (Claude Code, Cursor and VS Code docs); clients add fields often. It can't tell whether a server itself works; start it once to confirm. Unknown extra fields are left alone.