Workspace-mcp is a read-only tool suite with clear, consistent naming and solid descriptions. All 8 tools follow verb_noun patterns (list_, get_, read_) and have 50-100 character descriptions explaining purpose. However, most tools accept zero parameters (6 of 8), limiting schema complexity. The two parameterized tools (get_contract, read_repo_file) have proper type definitions and descriptions. No output schemas are documented, and error handling is minimal, tools catch broad exceptions but provide no recovery guidance. The server is well-engineered (strict mypy, ruff audit, pre-commit hooks) but lacks LLM-optimized error messages and pagination support for potentially large results.

• Document output schemas for all tools. Example: list_repos returns {repos: [{name: string, role: string}], total: number}. This enables LLMs to chain tools and extract the right fields.
• Add error recovery hints to tool descriptions. E.g., get_contract: 'If file not found, call list_contracts() to see available contract names.' This guides LLM self-correction.
• Add pagination to list_repos, list_contracts, and dependency_graph. Include limit (default 20, max 100) and offset/cursor parameters. Return total_count so LLMs know when to paginate.
• Enhance parameter descriptions with constraints. For read_repo_file.relpath, add: 'Relative path (no '..' or absolute paths). Max 500 chars. Alphanumeric, /, -, _ only.'
• Add tool annotations via fastmcp. Mark all tools with readOnlyHint=true and idempotentHint=true so LLMs know they are safe to retry without side effects.