A Go-based MCP server that provides tools for Docker image management, code searching, file conversion, Git operations, database management, and Kubernetes operations
Server has 7 tools with explicit schemas and descriptions, but quality is inconsistent. Tool names lack clear action verbs (e.g., 'to-markdown' should be 'convert_to_markdown', 'pull_image' should be 'pull_docker_image'). Descriptions are present but brief (avg ~40 chars, below 194-char baseline). Parameter descriptions exist but lack format/constraint details. No output schemas documented. Error handling returns generic text without recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Security concerns: tools execute arbitrary commands (exec, git_init, create_table) without input validation or sanitization against injection attacks.
Search for code in a file
Create a database table in a local Postgres DB instance
Get Kubernetes Pods from the cluster
Initialize a Git repository in the provided project directory
Run `mirrord exec` using a given config file
Pull an image from Docker Hub
Converts the provided input file to Markdown
No input validation or sanitization. Tools execute shell commands (markitdown, mirrord, git) with user-supplied paths and patterns. Vulnerable to command injection and path traversal attacks.
Tool names lack action verbs. 'to-markdown' should be 'convert_to_markdown'; 'pull_image' should be 'pull_docker_image'; 'mirrord-exec' should be 'exec_with_mirrord'. LLMs cannot infer intent from vague names.
No output schemas documented. Callers cannot predict response structure. E.g., get_pods returns unspecified format; create_table returns generic text. Breaks tool chaining.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | <=2025-11-25 | v2 |
Error responses are generic text without recovery guidance. E.g., 'failed to run markitdown: ...' does not tell LLM what to do next. No error classification (retryable vs fatal).
No tool annotations. Destructive tools (git_init, create_table, pull_image) lack destructiveHint. Read-only tools (ast-grep, get_pods) lack readOnlyHint. LLMs cannot reason about side effects.