An MCP server that helps generate comprehensive AGENTS.md files by reading project context, fetching documentation, and creating context summaries for AI agents.
The server defines 4 tools with explicit registration and Zod schemas. However, definition quality is significantly hampered by: (1) all tool descriptions contain a boilerplate 'NOTE' recommending use of the 'generate-agent-context' prompt, which is defensive and adds noise; (2) parameter descriptions are minimal and lack detail about expected formats, constraints, and error conditions; (3) no output schema documentation; (4) parameter names lack clarity (e.g., 'url' in read_docs could benefit from format documentation); (5) error handling returns generic text responses without structured guidance on recovery. The tools themselves are functional but fall short of LLM-optimized design. Average per-tool score: 42/100.
Create an AGENTS.md file with the provided context content. NOTE: For a complete automated workflow, it is recommended to use the 'generate-agent-context' prompt first.
Fetch documentation URLs for a list of packages using the AI client capabilities. NOTE: For a complete automated workflow, it is recommended to use the 'generate-agent-context' prompt first.
Get the project context from package.json, including dependencies, scripts, and other metadata. NOTE: For a complete automated workflow, it is recommended to use the 'generate-agent-context' prompt first.
Read the content of a documentation page from a URL. NOTE: For a complete automated workflow, it is recommended to use the 'generate-agent-context' prompt first.
All tool descriptions contain boilerplate 'NOTE: For a complete automated workflow, it is recommended to use the generate-agent-context prompt first.' This defensive tone and repetition across all 4 tools adds noise and suggests the tools are not designed for independent use. Descriptions should explain WHEN and WHY to use each tool standalone.
Parameter descriptions are minimal and lack actionable detail. Example: 'projectPath' has description 'Path to the project root (optional, defaults to current working directory).' but does not explain what format is expected (absolute vs relative), what happens if the path is invalid, or what the function does with missing package.json. LLMs need explicit guidance.
No output schema documentation. Tools return CallToolResult but the actual structure of the returned data (fields, types, examples) is not documented. LLMs cannot plan downstream tool calls or extract data reliably without knowing what get_project_context actually returns in 'content'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 45 | - | v1 |
read_docs accepts a 'url' parameter with format:'uri' constraint, but the description does not explain what types of URLs are supported, rate limits, timeout behavior, or what markup/text formats are returned. fetch_docs likely returns URLs, but chaining semantics are undocumented.
Error handling is generic. createAgentContextHandler catches errors and returns 'Error creating AGENTS.md: {error.message}' without distinguishing retryable errors (disk full, permission denied) from user-fixable errors (invalid targetPath). LLMs cannot determine next steps.
fetch_docs leverages Google Generative AI to find documentation URLs, but the tool description does not mention this dependency, explain what API key is required, or describe potential hallucinations in URL generation. This is a critical behavioral detail missing from the description.
create_agent_context has an optional 'targetPath' parameter, but the description does not clarify whether the path must be an absolute directory, whether it will be created if missing, or what permissions are required. Ambiguity invites failures.