64 MCP Tools

Full-text BM25F search with auto-grouping, relevance-gap filtering, and 30K token hard cap. The most-used CodeSift tool across 188 real sessions.

Ranked symbol discovery with BM25F scoring, kind filtering, and three detail levels. Finds function definitions, not string occurrences.

Nine built-in patterns for common code quality issues plus custom regex. Finds empty-catch, any-type, console-log, await-in-loop, and more.

Returns the full list of built-in pattern names with descriptions. Trivial token cost (~226 tokens).

Embedding-based code search for intent queries like 'error handling' or 'auth flow'. Supports Voyage AI, OpenAI, and Ollama providers.

Directory structure with symbol counts per file. First call in any unfamiliar codebase.

Structured summary of a file: exported symbols, types, signatures, line numbers. No implementation details.

Directories with file counts, symbol distribution by kind, top-level module summary.

Suggested queries based on indexed codebase. Top files by symbol density, kind distribution, example searches. ~240 tokens.

Retrieves a single symbol by ID with full source, metadata, and relationships. The precise second step after search_symbols.

Fetch multiple symbols by ID in a single call. More efficient than sequential get_symbol calls.

Find a symbol by name and show its body with optional references. Replaces grep + read in one structured call.

Returns a symbol plus its imports, siblings, and types. Everything needed to understand a symbol in context without reading the full file.

Where is X used? Distinguishes code references from comments, string occurrences, and import aliases.

Batch multi-modal queries: text, symbols, patterns, semantic, hybrid, conversation. Token budget mechanism fills context with highest-signal results.

Given a topic, assembles a complete context bundle with relevant signatures, types, entry points, and import relationships. Four compression levels from L0 to L3.

Build call graph for a symbol to configurable depth. Direction: callers or callees. Supports Mermaid output.

Trace HTTP route from handler to service to DB in one call. Supports NestJS, Next.js, Express. Optional Mermaid flowchart.

Given a git ref, identifies changed symbols, blast radius, affected tests, and risk scores. Answers 'is it safe to ship?'

Maps module dependency graph: imports, centrality scores, circular dependencies. Always use with focus parameter.

Louvain community detection on import graph. Finds natural module groupings that often don't match folder names.

Define dependency rules and check them against the import graph. Returns violations with file paths.

Classifies symbols as Hub, Bridge, Leaf, or Sink by call graph position. Reveals architectural risk.

Detect circular dependencies in the import graph via DFS. Returns file-level cycles with full import chain.

LSP-precise definition lookup understanding scope, imports, re-exports, and type aliases. Falls back to AST when LSP unavailable.

Return types, parameter types, documentation via LSP hover. 50-200 tokens instead of reading entire files.

Cross-file type-safe rename via LSP. Understands scope, shadowing, overloads, interface implementations.

LSP-backed incoming and outgoing call hierarchy for a symbol. Complements trace_call_chain with type-safe precision.

Ranks functions by cognitive complexity — nesting depth, not just branch count. Returns top N most complex functions, ranked with scores.

Git churn × complexity = risk rank. Files that change frequently AND are complex predict where bugs appear.

Hash bucketing and line similarity scoring to find structurally similar code. Detects same algorithm with different variable names.

Finds exported symbols with zero external references. Framework-aware whitelisting. False positive rate under 10%.

Clusters code by AST shape — finds patterns in how code is written, not just what it says.

Query the AST directly using tree-sitter syntax. Find code by shape, not text. Things impossible with regex.

Find imported names never referenced in the file body. Complements find_dead_code for import-level cleanup.

TruffleHog-derived rules with tree-sitter AST integration. Masked output, severity filtering, inline allowlist. The most dramatic benchmark result.

Shows functions/classes added, modified, or deleted between git refs. Semantic clarity over raw stat output.

AST-level symbol changes: renamed, signature changed, deleted. Essential for API compatibility.

Run 9 parallel static analysis checks on a git diff: secrets, breaking changes, coupling, complexity, dead-code, blast-radius, bug-patterns, test-gaps, hotspots.

Search symbols across all indexed repos simultaneously. Supports repo_pattern filtering.

Find all references to a symbol across all indexed repositories in one call.

BM25F + semantic search over past AI coding sessions. Find solutions discussed yesterday without re-debugging today.

Same as search_conversations but searches across all indexed projects simultaneously. Cross-project institutional knowledge.

Find all past sessions that discussed, modified, or analyzed a specific code symbol. Cross-references code with conversations.

Index Claude Code conversation history. Auto-discovery at startup, session-end hook for re-indexing. Noise filtering built in.

Create multi-step analysis plans with shared scratchpad, step dependencies, and automatic status tracking.

Get the current state of an analysis plan including all step statuses and results.

List all active analysis plans with their completion status and step counts.

Update step status in a plan. Automatically propagates to plan-level status on completion.

Write key-value data to a plan's scratchpad for cross-step knowledge sharing.

Read a key from a plan's scratchpad. Returns the stored value or null if not found.

List all entries in a plan's scratchpad with their sizes for inspection.

Search the full tool catalog by keyword or category. Returns matching tools with descriptions and parameter lists.

Get full parameter schemas for specific tools by name. Optionally reveal hidden tools in the active tool list.

Index a local folder with mtime-based incremental skip. Full reindex 57s → incremental 10s (5.6× faster).

Re-index one file after editing. 9ms unchanged, 153ms changed. Enables real-time coding workflows.

Clone a remote git repository and index it. Useful for onboarding or analyzing dependencies.

Returns all indexed repositories with metadata. Call once per session — result doesn't change.

Clears index for a repository, forcing full reindex. For major restructuring that mtime skip might miss.

Aggregate view of tokens consumed, tokens saved vs native baseline, and cost savings.

Generates CLAUDE.md from indexed codebase. Project structure, entry points, conventions, commonly used symbols.

One call, full codebase health picture: complexity hotspots, dead code, security, clones, communities, churn rankings.