codeloupe-mcp

Run a short sequence of codeloupe-mcp tools in one request when later calls depend on earlier results. Uses a deadline and fair output budget so earlier large results do not hide later calls; maximum 20 calls and recursive batch_tool_call is rejected.

Compare two source directory trees to quickly identify user-visible code changes. Use this before detailed reads when reviewing generated projects, migrations, releases, or copied workspaces; it reports added/deleted/modified/renamed files, bounded diffs, hotspots, and risk hints while skipping common generated/vendor directories by default.

Resolve one symbol on each side and return metadata plus a unified diff. Use after compare_directories or text_search when you need to review how a function/type changed across files or versions.

Report Tantivy content-index zone status for specific files or directories. Use before repeated text_search calls in large repositories to see whether scoped paths are ready, warming, not indexed, or outside the indexed workspace.

Rewrite one text file with normalized encoding and line endings. Use for cleanup before edits or tests; do not use for binary files.

Count lines in one file with encoding and binary detection. Use before choosing read_file_range/snippet bounds or estimating file size without loading content.

Create one directory with optional parent creation and structured success/error metadata. Use before creating files in a new path.

Create or overwrite one file with optional parent creation, encoding, and line endings. Use for new focused files; prefer edit_file for existing targeted changes.

Delete one file and return structured success/error metadata. Use only for explicit user-requested cleanup; directories are rejected.

Edit one text file using replace, append, prepend, or exact find-replace. Prefer find_replace with expected_replacements for surgical edits; use replace only when rewriting the whole file is intended.

Apply exact find-replace edits to up to 20 existing text files. Preflights every path, replacement count, and supported syntax before writing; writes each file atomically and rolls back earlier files if a later write fails. This is rollback-backed, not a filesystem-wide atomic transaction.

Compute the SHA-256 digest of one file with bounded-memory streaming I/O. Use the returned sha256 value as expected_hash for guarded edits or deletes.

Return quick metadata, binary detection, and a short preview for one file. Use as a cheap preflight before reading large or unfamiliar files.

Find symbol definitions with Tree-sitter for Rust, JavaScript/TypeScript, Python, C, C++, Go, Java, C#, PHP, Ruby, Swift, and Objective-C. Unsupported code extensions use a case-sensitive regex fallback labeled as heuristic.

Find token matches for a symbol across code files. Rust, JavaScript/TypeScript, Python, C, C++, Go, Java, C#, PHP, Ruby, Swift, and Objective-C matches are classified with Tree-sitter as code, comment, or string; unsupported code extensions use an explicitly labeled text fallback, not semantic reference resolution.

Perform fast fuzzy path and file-name search using the metadata/path index when available. Use before text_search to discover precise directories/files in large repositories; prefer concrete basename/path tokens and scoped paths over broad prose.

List outbound calls made from one function or symbol using Tree-sitter. Use after reading a symbol to understand direct dependencies without scanning the whole file.

Extract top-level symbols from one source file using Tree-sitter. Use after locating a file to choose exact symbols for read_symbol_body or call graph inspection.

Inspect persistent index storage and optionally remove orphaned, expired, explicitly selected, or least-recently-used workspace indexes. Dry-run is the default; active indexes are protected by cross-process locks.

List public declarations from supported Rust, JavaScript/TypeScript, Swift, Objective-C, Python, Go, Java, or C# files. Python uses the public-name convention; Go uses exported identifiers.

List recent in-memory write snapshots without returning file contents. History is process-local and bounded by entry count and retained bytes.

List imports from one supported Rust, JavaScript/TypeScript, Swift, Objective-C, Python, Go, Java, or C# file.

List up to 1000 archive entries or read one file of at most 10 MiB without extracting it. Use for source bundles or release artifacts; prefer inner_path for targeted reads.

Build a bounded tree view of a directory with optional size metadata. Use first for repository orientation and to choose precise text_search scopes; keep depth/children low for large repos.

Read one focused file range with encoding detection, full-file SHA-256, and truncation metadata. Use after search/path discovery; prefer line ranges over whole-file reads.

Read multiple focused file ranges in one request. Use to gather evidence from several files or call sites without loading whole files.

Read one symbol body with AST-first resolution, then heuristic fallback for other code-like files. Use when a function/type name is known and you need focused implementation context.

Normalize one path and return preflight accessibility metadata. Use when a user-supplied or relative path may be ambiguous before read/write operations.

Run fuzzy path, symbol-definition, and exact text searches concurrently. Results stay grouped by engine with explicit source, strategy, completeness, and limit status; one engine failure does not discard the other groups.

Check server uptime, workspace roots, and indexing health. Use before broad searches in large repos: each index_workspaces entry has a path_index (powers fuzzy/path tools) and a content_index (powers text_search shortlisting and may only cover listed zones).

Search file contents with exact literal or regex verification. Literal queries use Tantivy only for shortlisting and grep for correctness. Inspect search_strategy, fallback_reason, zone age, and unindexed-file diagnostics.

Restore the before-snapshot of one process-local history entry. Refuses to overwrite when the current path no longer matches the recorded after-snapshot.

Validate JSON syntax from one file or inline content without materializing the document. Provide exactly one of path or content; JSON Schema validation is not performed.

Request Tantivy content-index warming for specific scoped paths. Use before repeated literal text_search in large repositories; pass subsystem directories, then inspect statuses/warming_zones and retry when ready.

Estimate, enable, or disable indexing for one workspace. blocked paths and scan/disk budgets cannot be overridden. approval_required paths may be enabled only when the user explicitly requested work at that exact path. Otherwise use enable only for a user-relevant workspace when repeated repository searches are expected.

Summarize file, line, and language counts for a directory. Use to estimate repository size and decide whether text_search needs narrower paths/includes before content search.