mcp-config-check

# mcp-config-check [](https://github.com/MukundaKatta/mcp-config-check/actions/workflows/ci.yml) [](https://pypi.org/project/mcp-config-check/) [](https://pypi.org/project/mcp-config-check/) [](LICENSE) A small Python linter for [Model Context Protocol](https://modelcontextprotocol.io/) config files used by **Claude Desktop, Cursor, Cline, Windsurf, and Zed**. It catches common footguns before you start a broken MCP server: - missing or conflicting transport (no `command` and no `url`, or both set) - malformed `command` / `args` / `env` shapes - hardcoded API keys in `env` or `args` (Anthropic, OpenAI, AWS, GitHub, Stripe, Slack, Google, HuggingFace, and more) - placeholder values left in (`"<your-api-key>"`, `"replace-me"`, etc.) - URLs with embedded credentials (`https://user:pass@...`) - auth headers sent over plain HTTP - `autoApprove: ["*"]` / `alwaysAllow: ["*"]` wildcards that silently disable tool confirmation - case-insensitive duplicate server names - unknown fields in a server entry Supports both the Claude/Cursor/Cline/Windsurf shape (`mcpServers`) and the Zed shape (`context_servers`). ## Install ```bash pip install mcp-config-check ``` ## Usage ```bash mcp-config-check path/to/mcp.json ``` Multiple files: ```bash mcp-config-check ~/Library/Application\ Support/Claude/claude_desktop_config.json ~/.cursor/mcp.json ``` Only show errors (no OK lines, no warnings): ```bash mcp-config-check --quiet path/to/mcp.json ``` Exit status: `0` on no errors, `1` on any errors. ## Use as a library ```python from mcp_config_check import validate_config_file result = validate_config_file("path/to/mcp.json") if not result.ok: for issue in result.errors: print(issue.code, issue.server, issue.message) ``` ## Issue codes | Code | Severity | Meaning | |------|----------|---------| | E000 | error | file-level problem (missing, empty, not JSON, wrong root shape) | | E001 | error | server has no transport (`command` or `url` required) | | E002 | error | server declares both `command` and `url` | | E003 | error | `command` is not a non-empty string | | E004 | error | `args` is not an array of strings | | E005 | error | hardcoded secret detected in `args` | | E006 | error | `env` is not a string-valued object | | E007 | error | hardcoded secret detected in `env` value | | E008 | error | `env` value is an obvious placeholder | | E020 | error | `url` is invalid or has a non-http(s) scheme | | E021 | error | `url` has embedded credentials | | E022 | error | Authorization header sent over plain HTTP | | E030 | error | no `mcpServers` or `context_servers` container found | | E031 | error | case-insensitive duplicate server name | | E100 | error | `autoApprove` / `alwaysAllow` contains `"*"` | | W030 | warning | server container is empty | | W900 | warning | unknown field on a server entry | ## Use as a GitHub Action Add this step to any workflow: ```yaml - uses: actions/checkout@v5 - uses: MukundaKatta/mcp-config-check@v1 with: paths: .mcp.json ``` Pass multiple files space-separated (e.g. `paths: .mcp.json .cursor/mcp.json`). The action runs the same checks as the CLI and fails the workflow on any errors. Inputs: `paths` (required), `quiet` (default `false`), `python-version` (default `3.12`). ## Development ```bash pip install -e '.[dev]' pytest ``` ## License MIT