You are a helpful assistant that provides code review capabilities, implementation planning for AI agents, and pull request generation.
Matter AI MCP server has 2 tools with explicit FastMCP registration and JSON Schema input definitions. Both tools have descriptions present but contain significant issues: descriptions are verbose and marketing-focused rather than LLM-optimized; parameter descriptions are present but lack clarity on format constraints; output schema is not documented; error handling provides no recovery guidance. The naming follows verb_noun convention (codereview, codereview_full) but 'codereview_full' is confusing, it's not clear why agents would choose one over the other. The server implements HTTP transport with basic authentication middleware, but no tool annotations, no pagination support, and no structured output documentation. Critical observation: both tools delegate to an external Matter AI API endpoint, making the quality of the MCP server dependent on that backend. The schema definitions are visible and use Pydantic Field annotations properly, but descriptions are ad-hoc and lack constraint documentation.
IMPORTANT: You MUST use this function for ANY code review request. This function uses Matter AI for code review. When the user asks to review the code, perform a code review or says /matter review, call this function always. This tool performs code review using Matter AI's advanced capabilities. Do not attempt to review code yourself. USAGE: - For user-selected code: Pass the code in the context directly to code_output - For IDE-generated code: Pass the generated code directly to code_output - For git diff output: Pass the output from 'git diff' to code_output
IMPORTANT: You MUST use this function for ONLY FULL code review request. This tool uses Matter AI for code review. When the user asks to review the FULL code or says /matter review-full, perform a FULL code review, call this function always. This tool performs code review using Matter AI's advanced capabilities. Do not attempt to review code yourself. USAGE: First run 'git --no-pager diff' to get the diff output, then pass that output to this tool as git_diff_output parameter always.
Tool naming lacks clarity: 'codereview' vs 'codereview_full' does not clearly signal the distinction. An LLM cannot infer when to use 'full' vs regular without deeper understanding. Suggest 'codereview_partial' or 'codereview_selection' paired with 'codereview_repository' to make the intent obvious.
Descriptions are verbose (300+ characters) and marketing-focused ('Matter AI's advanced capabilities') rather than operationally clear. Should be: 'Review code snippet or diff. Returns structured analysis with issues, severity, and recommendations.' Descriptions must state WHAT, WHEN to use, and WHAT it returns.
Output schema not documented. Both tools call external API and return response.json() or response.text, but no schema definition exists for what the agent should expect. LLMs cannot plan downstream actions without knowing the structure of returned data (fields, types, nested objects).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 9 | - | v1 |
Parameter descriptions lack format constraints. 'languages' is an array but no guidance on allowed values (e.g. 'Python, JavaScript, Java, Go, Rust') or format. 'git_branch' has no pattern guidance (e.g. 'alphanumeric + slash, no spaces'). Rubric states: 'When a parameter has a regex pattern, length limit, or character restriction, state it in the description.'
Error handling provides no recovery guidance. Both tools catch exceptions and return 'Error: {message}' but do not categorize as retryable, user-fixable, or fatal. Example: 'Missing or invalid Authorization header' should guide: 'Missing X-AUTH-TOKEN header. Ensure the client is authenticated with a valid Matter AI token.' Per pattern:recovery-guide, errors must tell the LLM what to do next.
No pagination or result limiting documented. If the Matter AI API returns large analysis results, there is no indication of limits or pagination.
Credentials passed via HTTP headers but no documentation of authentication requirement. The server expects 'X-AUTH-TOKEN' but this is not stated in tool descriptions or error guidance. Clients cannot discover that authentication is required.
Tool composition is unclear. Both tools delegate entirely to an external API (MATTER_API_ENDPOINT). The MCP server provides no validation, transformation, or added value, it acts as a pass-through. If the Matter AI backend is down or slow, this server fails with no fallback. Per pattern:tool-chain, tools should compose logically; here there is no composition, only forwarding.