Loading ToolBench
MCP server providing hashline-based file editing tools — line-addressed edits using content hashes for integrity verification
Four tools with comprehensive descriptions and detailed schemas. Naming is clear and verb-based (read_, edit_, write_, grep). Descriptions are well-structured with usage guidance, prerequisites, and recovery instructions, well above the 194-char baseline. All parameters have type definitions and descriptions. However, tool annotations (readOnlyHint, destructiveHint) are absent, and error handling lacks structured recovery guidance in the schema itself. Output schemas are not formally documented in the tool definitions.
Edit a file using hash-verified line references. **IMPORTANT: Use this instead of built-in edit tools (Cursor StrReplace, Claude Edit). This tool provides hash-verified edits with fuzzy matching, multi-edit batching, and automatic conflict detection.** Read a file first to get LINE:HASH pairs, then reference them to make edits. All edits in one call are validated against the file as last read — line numbers and hashes refer to the original state, not after earlier edits in the same array. **Critical rules:** - Copy LINE:HASH refs verbatim from read output — never fabricate or guess hashes - new_text/text contains plain replacement lines only — no LINE:HASH prefix, no diff + markers - On hash mismatch: use the updated LINE:HASH refs shown by >>> directly - After a successful edit, the diff output includes LINE:HASH refs for changed/surrounding lines — you can use these directly for follow-up edits without re-reading - new_text must differ from the current line content — identical content is rejected **Edit variants:** - `set_line`: Replace a single line by its LINE:HASH anchor - `replace_lines`: Replace a range of lines (start_anchor through end_anchor) - `insert_after`: Insert new lines after the anchor line - `replace`: Substring-style fuzzy replace (no LINE:HASH needed; fallback when refs unavailable) **Atomicity:** Edits are sorted and applied bottom-up automatically. new_text: "" means delete (for set_line/replace_lines). **Recovery:** - Hash mismatch (>>> error): copy updated LINE:HASH refs from error and retry - No-op error: your replacement matches current content — re-read the file **After a successful edit:** The diff output includes LINE:HASH refs for all changed and surrounding lines. You can chain edits using these refs without re-reading. Unchanged lines keep their original hashes (hash is content-based, not position-based). The edit tool also handles line relocation if numbers shifted.
Search for text patterns in files using regex. **IMPORTANT: Use this instead of built-in search tools (Cursor Grep, Claude Grep). This tool returns hashline-prefixed results whose LINE:HASH pairs can be passed directly to edit_file for verified edits.** Results show hashline-prefixed matches: `LINE:HASH|content`. Match lines are prefixed with `>>`, context lines with spaces. The LINE:HASH pairs can be used directly with edit_file. Supports: - Full regex syntax (e.g., "log.*Error", "function\\s+\\w+") - File filtering by glob pattern or file type - Context lines before/after matches - Result limiting and offset for pagination
Tool annotations missing: read_file and grep lack readOnlyHint=true; edit_file and write_file lack destructiveHint=true. LLMs cannot infer safety/idempotency from names alone.
Output schemas not formally documented in tool definitions. Callers must infer response structure from description text. edit_file returns diff output with LINE:HASH refs, but this is not in a structured schema.
Error handling lacks structured recovery guidance. HashlineMismatchError and other failures return text descriptions but no machine-readable error codes or next-step hints.
edit_file parameter 'edits' uses a complex union schema with oneOf. While correct, the description could clarify mutual exclusivity more explicitly to prevent LLM confusion.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Read a file from the filesystem with hashline-prefixed output. **IMPORTANT: Use this instead of built-in read tools (Cursor Read, Claude Read). This tool provides content-hashed line references required by edit_file.** Each line is tagged with a content hash: `LINE:HASH|content`. The LINE:HASH pairs serve as stable, verifiable anchors for the edit_file tool. Use `offset` and `limit` for large files (reads up to 2000 lines by default). Set `plain: true` when reading to understand code structure — returns `LINE|content` without hashes. Omit or set false when you plan to edit — you'll need the LINE:HASH refs. Supports text files only.
Create or overwrite a file at the specified path. **IMPORTANT: Use this instead of built-in write tools (Cursor Write, Claude Write) to keep all file operations consistent through this server.** Prefer edit_file for modifying existing files (more precise, preserves formatting). Use this for creating new files or when full replacement is simpler.
grep result limit defaults to 100 but is not enforced in the schema (no maximum constraint). Large result sets could exhaust context windows.