Multi-agent orchestration platform with bundled MCP servers for Proton Mail, Reddit, and MailerLite integrations
This MCP server exposes 22 tools for setup, authentication, and MCP server management. While tool names follow verb_noun conventions (get_setup_status, install_github_cli, etc.), descriptions are present but often generic. Critical gaps: (1) Most tools lack documented output schemas, responses are inferred from code but not formally declared in tool definitions. (2) Parameter descriptions exist but are minimal (10-30 chars), below the 72-char baseline for A+ tools. (3) No error handling guidance, tools return success/error booleans without recovery hints. (4) No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels (WRITE, DESTRUCTIVE). (5) Sensitive operations like set_claude_api_key accept secrets as parameters, violating secret-injection pattern. (6) No pagination or result limits documented for list_* tools. Average tool quality is C-grade; most tools score 35-50 individually.
Create a new MCP server configuration
Delete an MCP server configuration
Get running status of all MCP server processes
Resolve a specific bundled MCP server by key
Check Claude CLI authentication status. Runs `claude auth status` and returns parsed result. Use this to poll after starting login
Check Codex CLI authentication status. Reads ~/.codex/auth.json to determine if the user is logged in
Get a single MCP server with unmasked env values (for editing)
Secrets exposed as tool parameters: set_claude_api_key accepts 'api_key' as a parameter. Credentials in tool params are logged in traces and prompt history, violating secret-injection pattern. Move to server-side environment variable or vault injection.
Output schemas not documented: 22 tools lack formal response schema declarations. Code shows Pydantic models (SetupStatusResponse, InstallResponse, etc.) but these are not exposed in tool definitions visible to LLMs. LLMs cannot plan downstream calls without knowing return structure.
No tool annotations for risk levels: Tools marked WRITE and DESTRUCTIVE (install_*, delete_mcp_server) lack destructiveHint/idempotentHint annotations. LLMs cannot distinguish safe from dangerous operations without explicit hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Get current setup status for all tools. Returns status of GitHub CLI installation and authentication, Claude CLI installation and authentication, Homebrew availability, npm availability, and overall setup completion status
Attempt to install Claude CLI using npm. If npm is not available, returns manual installation instructions
Attempt to install Codex CLI using npm. If npm is not available, returns manual installation instructions
Attempt to install GitHub CLI using Homebrew. If Homebrew is not available, returns manual installation instructions
List all available bundled MCP servers with resolved paths
List all configured MCP servers (env values masked)
List available tools for an MCP server by calling tools/list
Poll for GitHub authentication completion. Call this periodically after starting login to check if the user has completed authentication on GitHub
Set Claude API key for authentication. The API key should start with 'sk-ant-'. This method stores the key for use with Claude CLI
Start Claude CLI OAuth login (opens browser). Runs `claude auth login` which opens the user's default browser. Poll /claude-auth-status to detect when authentication completes
Start Codex CLI OAuth login (opens browser). Runs `codex login` which opens the user's default browser. Poll /codex-auth-status to detect when authentication completes
Start GitHub CLI login via device flow. Returns a one-time code that the user enters on GitHub. Opens the browser automatically to https://github.com/login/device
Test an MCP server configuration without persisting it
Update an existing MCP server configuration
Verify Claude CLI authentication by running a real subprocess. Runs a trivial prompt through Claude CLI to confirm authentication works. On success, writes a marker file for fast future checks
Minimal parameter descriptions: Most parameters have 10-30 character descriptions (e.g., 'Bundled server key identifier', 'MCP server name'). Baseline for A+ tools is 72 chars. Descriptions lack format constraints, valid ranges, or dependency hints.
No error recovery guidance: Tools return success/error booleans without actionable next steps. E.g., install_github_cli returns {success, error, manual_instructions} but no guidance on when to retry, what to check, or how to resolve. Agents cannot self-correct.