Index a repository and generate wiki documentation. This parses all source files, extracts semantic code chunks, generates embeddings, and creates wiki markdown files.

No prior indexing required.

Example: {"repo_path": "/path/to/repo"}

Ask a question about an indexed repository using RAG. Returns an answer based on relevant code context.

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "question": "How does authentication work?"}

Perform deep research on a codebase question using multi-step reasoning. Unlike ask_question (single retrieval), this performs query decomposition, parallel retrieval, gap analysis, and comprehensive synthesis. Best for complex architectural questions. Supports checkpointing for long-running research that can be resumed if interrupted.

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "question": "How is the event system architected?"}

Get the table of contents and structure of a generated wiki.

Requires: index_repository must be called first.

Read a specific wiki page content.

Requires: index_repository must be called first.

Semantic search across the indexed codebase with optional fuzzy matching and filters. Returns relevant code chunks with similarity scores.

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "query": "error handling"}

Export wiki documentation to static HTML files. Creates a self-contained website that can be viewed without a server.

Requires: index_repository must be called first.

Export wiki documentation to PDF format. Creates a printable PDF document with proper formatting, page numbers, and table of contents.

Requires: index_repository must be called first.

List all research checkpoints for a repository. Shows incomplete and cancelled research sessions that can be resumed using the deep_research tool with resume_research_id.

Requires: index_repository must be called first.

Cancel an active deep research session and save its checkpoint. The research can be resumed later using the deep_research tool with resume_research_id.

Requires: index_repository must be called first.

Resume a previously interrupted deep research session from its checkpoint. This is a convenience wrapper - you can also use deep_research with resume_research_id directly.

Requires: index_repository must be called first.

Get progress for active long-running operations. Supports polling-based progress tracking for clients that cannot receive push notifications. Returns current progress, ETA, and phase information.

No prior indexing required.

Get a searchable glossary of all code entities (classes, functions, methods) in an indexed repository. Useful for discovering what's in the codebase.

Requires: index_repository must be called first.

Generate Mermaid diagrams for an indexed repository. Supports class diagrams, dependency graphs, module overviews, language distribution pie charts, and sequence diagrams.

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "diagram_type": "class"}

Get class inheritance hierarchy trees for an indexed repository. Shows parent-child relationships, abstract classes, and generates a Mermaid inheritance diagram.

Requires: index_repository must be called first.

Get function call graphs showing which functions call which. Can analyze a specific file or the entire repository. Returns a Mermaid flowchart.

Requires: index_repository must be called first.

Get documentation coverage report for an indexed repository. Shows which classes, functions, and methods have docstrings and which don't.

Requires: index_repository must be called first.

Find wiki pages that may be outdated because their source files have been modified since the documentation was generated.

Requires: index_repository must be called first.

Extract recent git commit history as a formatted changelog. Groups commits by date and includes file change information.

No prior indexing required.

Scan a repository for hardcoded credentials and secrets (API keys, tokens, passwords, private keys). Returns findings with type, location, confidence, and remediation advice.

No prior indexing required.

Find usage examples for a function or class by searching test files in the indexed repository. Returns code snippets showing how the entity is used in tests.

Requires: index_repository must be called first.

Get API documentation with function signatures, parameters, return types, and docstrings for a specific file. Uses tree-sitter AST parsing for accuracy.

No prior indexing required.

Discover all indexed repositories under a given directory. Searches for .deepwiki directories and returns index metadata for each.

No prior indexing required.

Get index statistics for a repository without re-indexing. Shows file count, chunk count, languages, and when it was last indexed.

Note: This is an alias for get_status with scope='index'.

Requires: index_repository must be called first.

Full-text search across wiki pages and code entities. Searches titles, headings, code terms, descriptions, and keywords. Returns ranked matches.

Requires: index_repository must be called first.

Get parsed project metadata from package manifest files (pyproject.toml, package.json, Cargo.toml, go.mod, etc.). Returns name, version, dependencies, scripts, tech stack summary.

No prior indexing required.

Get rich context for a source file: imports, callers (who uses this file), related files, and type definitions used. Helps understand a file's role in the codebase.

Requires: index_repository must be called first.

Fuzzy name matching for functions, classes, and methods using Levenshtein distance. Returns 'Did you mean?' suggestions, file locations, and similarity scores. Great for finding entities when you don't know the exact name.

Requires: index_repository must be called first.

Get repository index status and/or wiki health dashboard.

• scope='all' (default): Returns both index status and wiki stats.

• scope='index': Index stats only (file count, chunks, languages).

• scope='wiki': Wiki health dashboard (pages, coverage, staleness).

Requires: index_repository must be called first.

Get a wiki health dashboard with index stats, page counts, search index size, coverage data, and wiki status - all in a single call.

Note: This is an alias for get_status with scope='wiki'.

Requires: index_repository must be called first.

Get a comprehensive explanation of a function, class, or method by combining glossary info, call graph, inheritance tree, test examples, and API docs into a single response.

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "entity_name": "MyClass"}

Analyze the blast radius of changes to a file or entity. Combines reverse call graph, inheritance dependents, file-level imports, and affected wiki pages to help understand impact before making changes.

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "file_path": "src/auth.py"}

Analyze code complexity for a source file using tree-sitter AST parsing. Returns function/class counts, line metrics, cyclomatic complexity, nesting depth, and parameter counts.

No prior indexing required.

Analyze git diff between two refs. Supports two modes:

• mode='structured' (default): Map changed files to affected wiki pages and code entities. Returns structured analysis.

• mode='question': Ask questions about the diff using RAG. Combines git diff with vector search context and LLM synthesis. Requires 'question' parameter.

No prior indexing required.

Ask questions about recent code changes using RAG. Combines git diff with vector search context and LLM synthesis to answer questions like 'What changed?', 'Are there any bugs?', or 'What's the impact?'.

Note: This is an alias for analyze_diff with mode='question'.

No prior indexing required.

Analyze architectural layer dependencies in a Python codebase. Categorizes files into layers (web, handlers, services, generators, core, providers, models) and detects upward dependency violations where lower layers import from higher layers. Returns layer file counts, dependency edges, and violations.

No prior indexing required.

Deprecated: use get_architecture_health with detail_level='full' instead. Returns a composite architecture overview combining health grade, layer dependency analysis, and file metrics.

No prior indexing required.

Rank functions across an entire repository by a chosen complexity metric (cyclomatic complexity, parameter count, line length, or nesting depth). Returns top-N hotspots with full detail breakdown. Useful for prioritising refactoring efforts.

No prior indexing required.

Build an inter-module import graph for a Python repository. Returns module nodes (with file counts and line counts), weighted directed edges, most-depended-on and most-dependent modules, and a Mermaid graph LR diagram.

No prior indexing required.

Compute Robert C. Martin package-level coupling metrics per module: afferent coupling (Ca), efferent coupling (Ce), instability (I = Ce/(Ca+Ce)), abstractness (A = abstract_classes/total_classes), and distance from the main sequence (D = |A+I-1|). Modules with high distance are either too concrete-and-stable or too abstract-and-unstable.

No prior indexing required.

Detect common design smells using heuristic AST-based thresholds: God Class (>15 methods AND >500 lines), Long Method (>80 lines AND cyclomatic complexity >7, or CC >15), Long Parameter List (>6 params), Feature Envy (>3 calls to another class's methods), Large File (>800 lines), Deep Nesting (>4 levels), Data Clump (3+ functions share 3+ identical parameter names). Returns smells with severity, file location, entity name, description, and refactoring suggestion.

No prior indexing required.

Comprehensive architecture health check. Runs complexity hotspot analysis, coupling metrics, design smell detection, and layer dependency analysis in a single call. Returns an overall health grade (A-F), per-dimension scores, and top findings.

No prior indexing required.

Compare architecture health between two git refs. Shows which metrics improved or degraded, grade changes, and new/resolved smells. Uses git worktree for safe non-destructive analysis.

No prior indexing required.

Comprehensive architecture analysis in a single call. Runs health check, dependency analysis, design smell detection, and hotspot ranking, then returns a pre-synthesized markdown narrative report. No prior indexing required.

Generate a developer onboarding guide for a codebase. Returns a markdown narrative with project overview, getting started instructions, repository layout, entry points, key modules, and testing info. No prior indexing required.

Generate prioritized refactoring recommendations from architecture health analysis. Returns actionable suggestions with effort/impact scoring. Set enrich=true for LLM-generated detailed descriptions. No prior indexing required.

Deep health analysis of a single module. Shows complexity distribution, design smells, coupling metrics, dependents (who uses this module), dependencies (what it uses), and refactoring risk level.

No prior indexing required.

View architecture health score trends over time. Returns historical snapshots with overall and per-dimension scores. Snapshots are saved automatically by 'deepwiki check' and 'deepwiki update'. No prior indexing required (reads saved history).

Generate a guided tour of a codebase organized by topic. Returns an ordered list of file stops with explanations. Topics: architecture, data_flow, request_handling, testing, or custom:. No prior indexing required.

Analyze file change frequency from git history. Shows which files change most often, optionally overlaid with cyclomatic complexity to find high-risk hotspots (frequently changed + complex). Also reports churn concentration (Gini coefficient).

No prior indexing required.

Find files that frequently change together in the same commits. Uses Jaccard similarity to measure co-change strength. High co-change coupling may indicate hidden dependencies or candidates for merging/splitting.

No prior indexing required.

Analyze class and module cohesion. Reports LCOM4 (Lack of Cohesion of Methods) for each class and internal-import ratio for each package. Classes with LCOM4 > 1 may be candidates for splitting. Modules with low cohesion ratio rely heavily on external imports.

No prior indexing required.

Detect code duplication using two methods: exact clones (Type 1, line-based fingerprinting) and structural clones (Type 2, AST node-type sequences). Reports duplication ratio, clone groups, and largest clone blocks.

No prior indexing required.

Analyze testability metrics for a Python repository. Reports test-to-code ratio, matches test files to source modules, counts assertions per test, and identifies untested source files.

No prior indexing required.

Compute Maintainability Index (MI) per function across a Python repository. MI combines Halstead Volume, cyclomatic complexity, and lines of code into a 0-100 score. Returns the worst-scoring functions first. Functions with MI < 20 are hard to maintain.

No prior indexing required.

Scan a repository for potential bugs using AST pattern matching. Detects bug patterns across Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, C#, and Kotlin including mutable default arguments, bare excepts, unreachable code, empty catch blocks, missing breaks, and more. Set enrich=true for LLM verification.

No prior indexing required.

Generate a Windsurf-style codemap: a focused execution-flow map with Mermaid diagram and narrative trace for a question or topic. Shows how code flows across files with file paths and line numbers. Best for understanding 'How does X work?' questions.

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "query": "How does request handling work?"}

Suggest interesting codemap topics for a repository based on call graph hubs, core modules, and common entry patterns. Use before generate_codemap to discover what flows are worth exploring.

Requires: index_repository must be called first.

Suggest which tools to use next based on tools already used. Returns ranked suggestions with reasons. No LLM calls - uses a static decision tree for instant responses.

No prior indexing required.

Run a pre-built multi-step workflow. Available presets: 'onboarding' (project overview), 'security_audit' (secrets + complexity), 'full_analysis' (stats + coverage + stale + secrets), 'quick_refresh' (stale docs + changelog).

Requires: index_repository must be called first.

Example: {"repo_path": "/path/to/repo", "workflow": "onboarding"}

Explain multiple code entities in a single call. Loads the search index once and looks up each entity name. More efficient than calling explain_entity repeatedly.

Requires: index_repository must be called first.

Smart query that combines ask_question with automatic escalation to deep_research when the initial answer is insufficient. Single entry point for codebase Q&A.

Requires: index_repository must be called first.

Search for tools by capability description. Returns ranked matches with tool name, description, and whether indexing is required. Useful when an agent needs to discover which tool to use.

No prior indexing required.

Start the interactive wiki web server for a .deepwiki directory. Launches a Flask app with chat, search, codemap explorer, and research UI. The server runs as a subprocess and can be stopped with stop_wiki_server.

Requires: index_repository must be called first.

Example: {"wiki_path": "/path/to/repo/.deepwiki"}

Stop a previously started wiki web server. Gracefully terminates the server process. If no server is found on the specified port, returns a list of currently running servers.

No prior indexing required.