DevLens

Codebase intelligence for TypeScript / JavaScript / React / Next.js / Node.js.
Turn any repo into a queryable graph of nodes and typed edges — each with AI technical, business, and security summaries — and use it from your CLI, your AI agent (MCP + Skill), or an interactive web UI. Runs entirely on your machine.

Join the Cloud Waitlist →

---

Click to watch the demo

---

Table of Contents

• What is DevLens?
• Why it helps
• The four surfaces
• Quick start
• 1 · CLI (@devlensio/cli)
  • CLI command reference
• 2 · MCP server
• 3 · Agent Skill (/devlens)
• 4 · Web UI
• Configuration
• What DevLens understands
• Architecture & repo layout
• DevLens Cloud
• Releasing
• Contributing
• License

---

What is DevLens?

AI coding tools let you ship faster than ever — but that speed creates a new problem: codebases grow faster than anyone (human or AI) can understand them. Developers merge code they didn't fully read; new hires drown in unfamiliar structure; agents burn tokens re-reading files they've seen before.

DevLens turns any React / Next.js / Node.js / TypeScript repo into a living, queryable map:

1. Walks the AST — extracts every component (with prop types), hook, function (typed params + return), store, utility, route, and file.
2. Builds a typed dependency graph — CALLS, IMPORTS, READS_FROM, WRITES_TO, PROP_PASS, EMITS, LISTENS, WRAPPED_BY, GUARDS, HANDLES, TESTS, USES.
3. Scores every node by architectural importance (multi-pass, no AI).
4. Summarizes each node with an LLM — a technical summary (what it does), a business summary (what it means for the product), and a security assessment (severity + notes).
5. Serves that graph to you through a CLI, an MCP server, an agent skill, and an interactive web UI.

The analysis engine is the separate devlensio package; DevLens OSS is everything you interact with on top of it. Everything runs locally — your code never leaves your machine.

---

Why it helps

• Understand AI-generated code — a visual + queryable map of what was built and how it connects, in plain English.
• Onboard in hours, not weeks — explore the graph and read per-node summaries instead of spelunking files.
• Blast-radius & impact analysis — before you change something, see exactly what depends on it.
• Architecture audits — surface central nodes, circular dependencies, and bottlenecks.
• Security review — every node carries a severity + explanation; list and prioritize them.
• Living documentation — summaries regenerate as code changes; unchanged nodes are reused (90%+ free on re-runs).
• Token-efficient agents — a node summary is ~50 tokens vs ~2000 for the file. Agents query the graph instead of reading files, and target work with blast-radius/k-hop instead of scanning directories.

---

The four surfaces

Surface	What it is	Install	Best for
CLI	devlens command — analyze + query from the terminal, JSON-friendly	npm i -g @devlensio/cli	scripts, terminals, CI
MCP server	devlens mcp — 14 tools over the Model Context Protocol	bundled in the CLI	Claude, Cursor, any MCP client
Agent Skill	/devlens — teaches an agent how to drive the CLI	npx @devlensio/skill install	Claude Code, Cursor, Kilo, opencode, pi
Web UI	interactive graph visualizer (Next.js)	bun run dev (from source)	visual exploration

All four read the same graph stored in ~/.devlens.

---

Quick start

Install

Standalone binary — one command, no Node or Bun required (downloads a prebuilt native binary from GitHub Releases into ~/.devlens/bin):

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/devlensio/devlensOSS/main/scripts/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/devlensio/devlensOSS/main/scripts/install.ps1 | iex

Override the location with DEVLENS_INSTALL_DIR, or pin a version with DEVLENS_VERSION=v0.2.6 ($env:DEVLENS_VERSION on Windows). The Windows script also adds the install dir to your PATH.

Or via npm (cross-platform, requires Node) — also bundles the MCP server:

npm install -g @devlensio/cli

Use

# One-time: pick an LLM provider for summaries (or skip for structure-only)
devlens init

# Analyze a repo (run inside it) — add --summarize for AI summaries
devlens analyze . --summarize

# Explore
devlens overview
devlens find-nodes Button
devlens blast-radius "src/auth/login.ts::login"

Behind a corporate proxy? If npm install fails with UNABLE_TO_VERIFY_LEAF_SIGNATURE, point Node at your org's root CA: export NODE_EXTRA_CA_CERTS=/path/to/corp-ca.pem.

---

1 · CLI (@devlensio/cli)

The devlens command analyzes a repo and lets you query the resulting graph. Every query command supports --json (machine-readable) and -g <graphId> / -c <commit> to target a specific graph/commit (default: the graph for the current directory).

Typical flow: devlens init → devlens analyze . --summarize → query commands (overview, find-nodes, get-node, blast-radius, …).

CLI command reference

Global options available on every command: --json (machine-readable output), -v, --verbose (diagnostics), -h, --help.

Setup & environment

Command	Description
devlens init	First-time setup — interactively configure your LLM provider for summarization.
devlens doctor	Check environment health (git, storage, LLM provider). Run this when something fails.
devlens config [options]	Show or update config in ~/.devlens/config.json.
devlens status	Show analyzed + summarized graphs (per graph: commits, latestCommit, summarizedCommits).
devlens repos	List repositories DevLens has already analyzed.
devlens graphs list / devlens graphs delete <graphId>	List or delete stored graphs.

devlens config options: --set (interactive), --provider <anthropic\|openai\|openrouter\|gemini\|ollama>, --model <m>, --api-key <k>, --base-url <u> (e.g. http://localhost:11434 for Ollama), --batch-size <n>.

Analyze & summarize

devlens analyze [path] [commitHash] — analyze a repo into a graph.

• path (default .) — repository path.
• commitHash — informational; the engine analyzes the working tree.
• --summarize — also generate technical/business/security summaries (costs LLM calls).
• --force-summarize — re-summarize every node from scratch.
• --latest — analyze the working tree including uncommitted changes (current default).
• Example: devlens analyze . --summarize

devlens summarize [target] [commit] — run analysis then summarization.

• target (default .) — repo path or an existing graphId.
• --force-summarize — ignore prior summaries.
• --model <model> / --provider <anthropic|openai|openrouter|gemini|ollama> — override for this run.

Orient & find

devlens overview — repo fingerprint (framework, router, state, data, db), stats (totalNodes/totalEdges), routeCount, and the most-central nodes/files. Start here. (opts: -g, -c)

devlens top-nodes — highest-scoring (most central) nodes. -l, --limit <n> (default 25). (opts: -g, -c)

devlens find-nodes [name] — search/filter nodes (compact refs, not source).

• name — substring match on node name.
• -t, --type <types...> — COMPONENT, HOOK, FUNCTION, STATE_STORE, UTILITY, ROUTE, FILE, TEST, STORY, THIRD_PARTY.
• -f, --file <path> — nodes in exactly this file. -d, --dir <path> — nodes under this folder.
• --node-ids <ids...> — fetch exact ids. --min-score <n>. --severity <low|medium|high> — min security severity.
• -l, --limit <n> (default 25). (opts: -g, -c)
• Example: devlens find-nodes -t ROUTE -l 500 --json

devlens nodes-in-path <path> — every node in a file or folder. -t, --type <types...> to filter. (opts: -g, -c)

Understand (summaries & detail)

devlens get-node <nodeId> — full detail for one node: summaries + callers + callees (each connection carries viaEdge). Your main inspection tool.

• -i, --include <sections...> — metadata|callers|callees|technical|business|security (narrow the payload).
• -e, --edge-types <types...> — filter caller/callee edges. (opts: -g, -c)

devlens get-summaries <nodeIds...> — batch summaries for many ids. -i, --include <kinds...> — technical|business|security (business = functional). (opts: -g, -c)

devlens node-code <nodeId> — raw source for a node. Expensive — prefer get-node. (opts: -g, -c)

Structure & impact

devlens blast-radius <nodeId> — upstream dependents ("what breaks if I change this"). -r, --radius <n> (default 2, capped on huge fan-out; an explicit value is uncapped). -e, --edge-types <types...>. (opts: -g, -c)

devlens khop <nodeId> — downstream dependencies ("what this needs"). Same -r / -e. (opts: -g, -c)

devlens subgraph <seedNodeId> — the cohesive cluster around a node (returns clusterId, nodes, edges). (opts: -g, -c)

devlens cycles — circular-dependency groups. (opts: -g, -c)

devlens diff <from> <to> — changed nodes between two commits + blast radius of the changes. -r, --radius <n> (default 1). Both commits must already be analyzed into the graph. (opts: -g)

Security

devlens security — nodes flagged with security concerns (severity + per-node explanation). --min-severity <low|medium|high> (default low). -l, --limit <n> (default 50). (opts: -g, -c)

Serve & integrate

devlens serve [path] — start the backend API server (used by the web UI). -p, --port <port> (default 3000).

devlens mcp — run the MCP server. Subcommands: stdio (default — for editors/agents) and http [-p, --port <port>] (Streamable HTTP, default port 7000). See §2.

---

2 · MCP server

The CLI binary embeds a Model Context Protocol server exposing 14 tools over the same graph (list_analyzed_repos, get_repo_overview, find_nodes, get_node, get_blast_radius, get_khop, get_subgraph, list_cycles, get_security_issues, analyze_changes, get_summaries, get_nodes_in_path, get_node_code, and more).

devlens mcp            # stdio (what an editor/agent spawns)
devlens mcp http -p 7000   # Streamable HTTP at http://localhost:7000/mcp

Register it in a client (Claude Code shown):

claude mcp add devlens -- devlens mcp

Or add to any MCP client config:

{ "mcpServers": { "devlens": { "command": "devlens", "args": ["mcp"] } } }

Windows + Claude Desktop: if the server won't start, wrap it: { "command": "cmd", "args": ["/c", "devlens", "mcp"] }.

Registered in the official MCP registry as io.github.devlensio/devlens.

---

3 · Agent Skill (/devlens)

The DevLens Agent Skill teaches your AI coding agent how to drive the CLI — when to query the graph, how to keep output token-cheap, and how to produce thorough architecture/security/impact reports. It's one /devlens command with subcommands.

# Install into whatever AI tools it detects (Claude Code, Cursor, Kilo, opencode, pi)
npx @devlensio/skill install

Or, for Claude Code, install the plugin:

/plugin marketplace add devlensio/devlensOSS
/plugin install devlens@devlensio

Subcommands: /devlens init, architecture, diagram, summary, security-analysis, explain, tech-debt, impact, find, changes, guard. See plugins/devlens/README.md and packages/skill-installer/README.md.

---

4 · Web UI

An interactive graph visualizer (Next.js + Cytoscape) — force-directed canvas, node detail panels, search/filter, commit-diff overlay, and a security panel. Runs from source against the local backend:

git clone https://github.com/devlensio/devlensOSS.git
cd devlensOSS
bun install
bun run dev          # starts the backend (src/server) + the frontend

Open the printed URL, paste the absolute path to a repo (must have a root package.json), and click Analyze.

---

Configuration

Config lives in ~/.devlens/config.json and is set via devlens init / devlens config (or the Web UI's Config panel). Summaries need an LLM provider; structural analysis does not.

Supported providers

Provider	Recommended model	Notes
Ollama (local)	qwen2.5-coder:7b	Free, local, 8GB+ RAM. Slow on large repos.
OpenAI	gpt-4o-mini	Fast, cost-effective
Anthropic	claude-haiku-4-5	Excellent code understanding
OpenRouter	grok-4.1-fast	100+ models; best cost/quality balance
Gemini	gemini-2.0-flash	Fast, large context

devlens config --provider openrouter --model grok-4.1-fast --api-key <key>
# Ollama (local, no key): devlens config --provider ollama --model qwen2.5-coder:7b --base-url http://localhost:11434

Avoid free-tier OpenRouter models — their rate limits are too low for a full summarization run.

---

What DevLens understands

Node types: COMPONENT, HOOK, FUNCTION, STATE_STORE, UTILITY, FILE, ROUTE, TEST, STORY, THIRD_PARTY (plus an internal GHOST placeholder).

Edge types: CALLS, IMPORTS, READS_FROM, WRITES_TO, PROP_PASS, EMITS, LISTENS, WRAPPED_BY, GUARDS, HANDLES, TESTS, USES.

Per node: an importance score, a technical summary, a business/functional summary, and a security assessment (none|low|medium|high + notes). Connection results carry the edge type via a viaEdge field.

---

Architecture & repo layout

DevLens OSS is the set of interfaces built on top of the devlensio analysis engine (a separate package).

devlensOSS/
├── src/
│   ├── cli/                 # the `devlens` CLI (commander program + commands)
│   ├── core/                # shared query core (used by CLI + MCP — never drift)
│   ├── mcp/                 # MCP server (stdio + Streamable HTTP), 14 tools
│   └── server/              # backend API for the Web UI (SSE job streaming)
├── frontend/                # Next.js 15 graph visualizer (Cytoscape)
├── bin/devlens.cjs          # npm launcher → execs the per-platform native binary
├── npm/<platform>/          # 5 prebuilt-binary packages (@devlensio/cli-<os>-<arch>)
├── plugins/devlens/         # the /devlens Agent Skill (Claude plugin)
├── packages/skill-installer/# @devlensio/skill — the npx skill installer
├── scripts/                 # set-version.mjs, set-skill-version.mjs, stage-binaries.mjs
├── .claude-plugin/          # marketplace.json (Claude plugin catalog)
└── server.json              # MCP registry manifest

The analysis pipeline (AST parsing, edge detection, scoring, summarization) lives in devlensio, not here.

---

DevLens Cloud

A hosted version is in development: GitHub integration (no local clone), shareable graphs, team annotations, a natural-language interface over your graph, semantic search across summaries, automatic PR review summaries, and persistent cloud storage.

Join the waitlist →

---

Releasing

The repo ships independently-versioned artifacts:

Artifact	Package	Version source	Channel
CLI + MCP	@devlensio/cli (+ 5 platform pkgs)	scripts/set-version.mjs	npm (Trusted Publishing / OIDC) on v* tag
Agent skill	@devlensio/skill + Claude plugin	scripts/set-skill-version.mjs	npm + git (see packages/skill-installer/README.md)
Engine	devlensio	the separate devlens-engine repo	npm

Release the CLI

scripts/set-version.mjs <ver> stamps every manifest in lockstep — the main package, all 5 npm/<platform>/package.json, the pinned optionalDependencies, server.json (MCP registry), and the CLI's hardcoded --version. Pushing a v* tag triggers .github/workflows/release.yml, which builds the binaries and publishes to npm (via OIDC), the MCP registry, and the GitHub release.

# 0. (if the engine changed) publish devlensio first, then bump the "devlensio" pin here + `bun install`
node scripts/set-version.mjs 0.2.6        # 1. bump all manifests
git add -A && git commit -m "release: @devlensio/cli 0.2.6"   # 2. commit (include bun.lock)
git push && git tag v0.2.6 && git push origin v0.2.6          # 3. tag → triggers CI

The binaries bundle whatever devlensio resolves to at build time; CI installs the published version from the pin, so an engine fix reaches users only after devlensio is republished and the pin is bumped. After bumping deps, run bun install so bun.lock matches (CI uses --frozen-lockfile).

Release the Agent skill

Versioned separately from the CLI. node scripts/set-skill-version.mjs <ver> stamps the installer + plugin together; then npm publish the installer and push the plugin. See packages/skill-installer/README.md.

---

Contributing

DevLens is actively under development. Bugs, ideas, and PRs welcome — open an issue or PR. All contributions remain open source under AGPL v3. A detailed CONTRIBUTING.md will follow as the project matures.

---

License

DevLens is licensed under the GNU Affero General Public License v3.0.

You are free to use, modify, and distribute DevLens. If you run a modified version as a hosted service, you must release your modifications under the same license.