KiroGraph is a semantic code knowledge graph for Kiro IDE — and a growing family of other MCP-capable AI tools. It parses your source code using tree-sitter, extracts every symbol and relationship, and stores the result in a local SQLite database. When Kiro needs to understand your codebase, it queries that graph in a single tool call instead of reading dozens of files one by one.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/davide-desio-eleva/kirograph/llms.txt
Use this file to discover all available pages before exploring further.
Why KiroGraph?
When an AI agent works on a complex task, it typically explores the codebase through file reads,grep, and glob searches. Every one of those is a tool call, and tool calls consume context and slow responses down. On a large feature or a tricky bug, a single task can trigger 10–30 such calls before the agent has enough context to act.
KiroGraph collapses that work into one. A single kirograph_context("fix auth bug") call returns:
- The symbols most likely relevant to the task
- Their call graph neighbors and type hierarchy
- The impact radius — what else would break
- Linked memory observations from past sessions (if enabled)
KiroGraph stores everything in
.kirograph/kirograph.db on your machine. No data leaves your project, no API keys are required, and no external services are contacted during indexing or querying.Architecture at a Glance
KiroGraph is built around a core graph layer that is always active, with opt-in modules you enable as your project needs grow.Core Graph (always on)
Tree-sitter AST parsing across 33+ languages extracts 26 node kinds — functions, classes, methods, routes, components, and more — plus the edges between them: calls, imports, exports, extends, implements, contains, references. Everything lands in a single SQLite file with no native dependencies required for structural indexing.Opt-in Modules
| Module | Config flag | What it adds |
|---|---|---|
| Semantic Search | enableEmbeddings: true | 768-dim vector embeddings via HuggingFace; 9 engine choices |
| Architecture | enableArchitecture: true | Package graph, layer detection, coupling metrics |
| Security | enableSecurity: true | Reachability-aware CVE scanning, secrets detection, SAST-lite, SBOM/SARIF |
| Memory | enableMemory: true | Cross-session observations auto-linked to code symbols |
| Watchmen | enableWatchmen: true | Synthesizes memory into workspace briefs and skill files |
| Wiki | enableWiki: true | Karpathy-style LLM wiki that compounds knowledge across sessions |
| Docs | enableDocs: true | Section-level retrieval from Markdown, MDX, RST, AsciiDoc, OpenAPI |
| Data | enableData: true | Query CSV, JSON, Excel, Parquet, and PDF data server-side |
| Token Optimization | shellCompressionLevel, cavemanMode | 60–90% token savings on shell output and agent prose |
Platform Support
Full support is for Kiro IDE. Experimental auto-detection and integration is available for 34 other MCP-capable tools, including Cursor, GitHub Copilot, Claude Code, Windsurf, Cline, Codex, Goose, Aider, Replit, and more.Requirements
- Node.js >= 18
- Kiro IDE (fully supported) or any other MCP-capable tool (experimental)
Where to Go Next
Quickstart
Install KiroGraph, run the installer, index your project, and make your first graph query in under five minutes.
How It Works
Dive into the indexing pipeline — tree-sitter parsing, the graph model, incremental sync, and how MCP tools query the graph.
Integrations
Set up KiroGraph for Kiro IDE or any of the 34 other supported MCP-capable tools.
MCP Tool Reference
Full reference for all MCP tools exposed by the KiroGraph server, organized by module.