A Go-based MCP server providing semantic code search, call graph analysis, and hybrid search capabilities for AI agents to understand and navigate codebases
Semantic-grep has 7 tools with mixed quality. Tool names follow verb_noun convention (search_*, edit, read_*, list_*, directory_tree), which is good. However, descriptions vary widely: search_memory and search_call_graph have solid 100+ char descriptions explaining purpose and parameters; others like list_directory (11 chars) and directory_tree (35 chars) are critically short. Input schemas are present for all tools with proper type definitions (string, integer, enum), but parameter descriptions are inconsistent, some detailed, others minimal. Output schemas are not documented anywhere in the source. Error handling is absent: no guidance on retryability, user-fixable vs fatal errors, or recovery paths. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY vs WRITE). The edit tool accepts file paths without sanitization, creating path traversal risk. Overall, the server reads like a basic wrapper around file/code operations without LLM-optimized design.
Generates a tree structure of a directory.
Edits the specified file with the given modifications. This tool marks files that need to be edited with the specified changes.
Lists the contents of a directory.
Reads the content of a text file at the specified path. You can optionally specify an offset and limit to read only a portion of the file.
Searches the call graph to find callers and callees of a given function symbol, with optional depth limit and direction control
Searches the semantic memory index for code chunks matching a query using hybrid search (vector similarity + keyword matching)
Searches for a regular expression pattern within the content of files in a specified directory (or current working directory). Can filter files by a glob pattern. Returns the lines containing matches, along with their file paths and line numbers.
These tools are underdocumented for LLM selection.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns. This violates the pattern:tool requirement.
No error handling guidance. Tools provide no recovery hints (e.g., 'file not found, check path' or 'search returned no results'). Agents cannot self-correct on failures.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
edit tool accepts file_path without sanitization or validation. No checks for path traversal (../../../etc/passwd), symlink attacks, or permission boundaries. Destructive operation (WRITE risk) lacks confirmation or dry-run support.
No tool annotations despite clear risk levels. search_* and read_* are marked READ_ONLY; edit is WRITE. Annotations (readOnlyHint, destructiveHint) enable agents to reason about side effects and retry safety.