mcpu exposes 9 tools with minimal descriptions and incomplete schemas. Tool names use repo.* and grant.* prefixes but lack action verbs (e.g., 'repo.status' instead of 'get_repo_status'). Most descriptions are generic placeholders ('repo.status for this Artifact-backed repo') under 50 chars, providing no context for LLM selection. Input schemas are present but sparse: most parameters lack descriptions (e.g., 'grant' param has description, but 'path', 'contents', 'message' lack detail on format/constraints). No output schemas documented. Error handling is minimal, tools throw generic errors without recovery guidance. The grant.create tool has the most complete schema with 4 parameters and descriptions, but others are skeletal. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels (WRITE, IRREVERSIBLE). Overall, definitions are functional but far below production quality.

• Rename tools to verb_noun pattern: 'get_repo_status', 'list_repo_files', 'read_file', 'write_file', 'get_repo_diff', 'commit_changes', 'deploy_artifact', 'get_repo_history', 'create_grant'.
• Expand descriptions to 100-200 chars. E.g., 'get_repo_status: Returns the current state of the artifact repository, including version, storage backend, draft status, and deployed artifact ID. Call this first to understand the repo state before making changes.'
• Add descriptions to all parameters. E.g., 'path: File path relative to repo root (e.g., src/index.ts). Must not contain .. or absolute paths.'; 'contents: File contents as UTF-8 string. Max 1MB.'
• Document output schemas in tool descriptions or as JSON Schema. E.g., 'Returns: {pack: string, version: string, storage: string, remote: string|null, draftDirty: boolean, currentArtifact: string|null, deployedArtifact: string|null}'.
• Add tool annotations: readOnlyHint=true for repo.status/ls/read/diff/history; destructiveHint=true for repo.deploy; idempotentHint=true for repo.read/status/history.