Local MCP server: verify repo changes (read, search, tests, lint) with a safe workspace boundary.
Server has 6 tools with explicit schemas and descriptions. Naming is mostly clear (verb_noun pattern: read_repo_file, search_codebase, run_tests, run_lint). Descriptions are present but vary in quality, some are detailed (solution_trace at ~400 chars), others minimal (verifier_health at ~80 chars). All tools have input schemas with type definitions. However, parameter descriptions are sparse or missing in several tools (e.g., run_tests and run_lint lack detail on what extra_args actually does beyond 'shell-split'). Output schemas are not explicitly documented, responses are JSON strings without declared structure. Error handling is present but generic (returns error in JSON payload without recovery guidance). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only operations.
Read a text file under the workspace root (read-only). Path must be relative; no absolute paths.
Run the project lint command (override with RULE_BASED_LINT_CMD). Optional extra_args are shell-split.
Run the project test command (override with RULE_BASED_TEST_CMD). Optional extra_args are shell-split.
Search the workspace with ripgrep when available, else a Python regex scan. Pattern is a ripgrep regex (or Python regex in fallback mode).
One entry point for solution graphs. **mode** (case-insensitive): - ``git`` (default): git diff → hunk nodes + Python import edges. Args: ``ref``, ``staged``, ``write_svg_relative``. Needs ``.git`` (mount full repo in Docker). - ``payload``: explicit JSON object with ``chunks`` and optional ``edges``. Arg: ``payload_json``. - ``semantic``: AST symbol add/remove/modify + call/import edges for changed ``.py`` files. Args: ``ref``, ``staged``, ``write_svg_relative``, ``write_html_relative`` (optional interactive HTML). All modes: Markdown + JSON; optional localhost SVG preview when ``serve_localhost`` is true.
Output schemas not documented. Tools return JSON strings but LLMs cannot see the structure of fields returned (e.g., what fields does search_codebase result contain?). This forces LLMs to guess at response structure and risks parsing errors.
Parameter descriptions are minimal or missing context. 'extra_args' in run_tests and run_lint only says 'shell-split' but does not explain what arguments are valid or when to use them. LLMs cannot infer valid values.
No tool annotations despite all tools being read-only. Adding readOnlyHint=true to all 6 tools would signal to agents that these are safe to call without side effects, improving planning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Report server version, resolved workspace root, and detected test/lint commands.
Error responses lack recovery guidance. When read_repo_file fails (file not found, too large, permission denied), the response includes the error but does not suggest next steps. Pattern: 'File not found. Try search_codebase() to locate it.'
solution_trace parameter 'mode' accepts free-form string with case-insensitive matching, but no enum constraint. LLMs may hallucinate invalid modes like 'diff' or 'ast'. Should declare enum: ['git', 'payload', 'semantic'].