DevLens

Codebase Intelligence for AI agents and Developers.

Analyze your repo once. Your AI agent then queries a precomputed code graph through MCP — ranked files, call flow, impact, and security — instead of re-reading your codebase file by file. Five self-describing MCP tools, plain-text packets measured at a third of the tokens of reading the code, and the smallest response in its benchmark field.

Join the DevLens Cloud Waitlist →

Click the image to watch the demo

Why DevLens?

• Your agent stops burning tokens. One MCP call returns a task-shaped packet — the files and symbols that matter, ranked, within a token budget. No grep loops, no re-reading files.
• It wins the benchmark. First of nine tools on correctness, F1, recall, and precision — the only tool that leads every quality column.
• It is fast and small. 67 ms per call, packets 6.7× smaller than the closest graph competitor, and it never exceeds the token budget you set.
• Developers get it too. Interactive graph visualization, blast radius before you change a symbol, PR review packets, and per-node security analysis.

Benchmarks

We benchmarked DevLens MCP against other code graph and retrieval tools — Graphify, codegraph, semble, bm25, and the grep-and-read floor — on one headless coding agent, one model, one real TypeScript repository, and 133 real questions written from actual pull requests and symbols. The current consolidated 5-tool surface (V3), same-run against the field:

DevLens V3 ties the strongest competitors at 43% fewer tokens than the old 24-tool surface, with the smallest packets in the field and the fastest calls (67 ms per call at the retrieval layer, ~37× faster than the next graph tools). Where no tool — including grep — wins, the multi-file synthesis questions, V3 leads the field on cost: feature-intent at 154k tokens vs the floor's 223k at equal correctness.

The pattern behind the numbers: syntactic tools return big blobs or bare line lists. DevLens returns a task-shaped packet from a type-resolved graph — the files that matter, ranked, within your token budget, in milliseconds.

Full methodology, per-type tables, and the nine-tool round: docs/PUBLIC-BENCHMARKS.md

How DevLens compares

DevLens is the only tool in this space that combines native semantic parsing, per-node AI summaries with per-node security analysis, and framework-aware data edges — and the only option you can use commercially under AGPL.

(Feature comparison from public sources, Aug 2026.)

Why teams choose DevLens: semantic edges that syntactic tools cannot produce, per-node security analysis no other open-source tool provides, benchmark-winning retrieval quality, and a single MCP front door — free for commercial use.

Quick Start

Step 1 — Install the CLI

npm install -g @devlensio/cli

Or install the standalone binary (no Node.js needed):

# Linux / macOS
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

Step 2 — Configure a model (optional)

Only needed if you want AI summaries. Structure-only analysis works offline with no API key.

devlens init

This walks you through picking a provider, API type, key, endpoint, and model interactively — press ESC at any step to go back a stage. See Configuration for recommended models.

Step 3 — Analyze your repo

cd your-project
devlens analyze . --summarize

This builds the graph, generates summaries, and creates the search index — once. After this, everything below is instant.

Step 4 — Connect your AI agent

Start the MCP server:

devlens mcp

Then register it in your agent. Pick your tool:

Claude Code / Claude Desktop

claude mcp add devlens -- devlens mcp

Cursor — add to .cursor/mcp.json:

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

Codex — add to ~/.codex/config.toml:

[mcp_servers.devlens]
command = "devlens"
args = ["mcp"]

Hermes — add to config.yaml:

mcp:
  servers:
    devlens:
      command: devlens
      args: [mcp]

Any other MCP client — the server speaks standard MCP over stdio. Point your client at the command devlens mcp. For an HTTP transport instead:

devlens mcp http -p 7000

The server self-describes its full tool list through tools/list, so your agent discovers everything on its own. That's it — ask your agent "how does auth work in this repo?" and watch it query the graph instead of reading files.

What your agent can ask

The server self-describes every tool's parameters through tools/list — your agent fills them in automatically. Five tools, no graphId anywhere — the repo is resolved from your working folder.

Every response carries a provenance block (commitHash, analyzedAt, hasGit), so your agent always knows which snapshot answered the question.

Full reference: docs/mcp-reference.md

Supported languages

DevLens parses five language families with native parsers (no AI, no regex, no tree-sitter) and understands their frameworks. Summaries are never generated silently: you confirm before tokens are spent, and structure-only analysis needs no provider at all.

Each repo is analyzed with its language's own parser (Python ast, JavaParser, Go go/ast plus go/types, Rust syn, TS compiler API), so edges are real: type-checked interfaces (IMPLEMENTS), framework routes (HANDLES), and ORM data layers (READS_FROM, WRITES_TO).

For developers: the Web UI

For when you want to see your codebase laid out as an interactive graph.

Open the Web UI, paste your repo path, and explore a force-directed canvas. Click any node to see its summaries, callers, callees, and security flags. Search, filter, and diff commits across versions. The Project panel shows search index readiness with a one-click reindex action.

bun start   # from the source repo: API + Web UI on one port (http://localhost:3000)

The Web UI runs from the source tree, it is not bundled into the installed CLI binary.

CLI

For scripts, CI, and fast answers without leaving the terminal. Every command supports --json for piping into scripts, -v/--verbose for diagnostics, and --quiet for minimal output.

Analyze and summarize

Explore and understand

Impact and quality

Manage and integrate

Hands-on examples

devlens analyze . --summarize             # graph + summaries + search index
devlens find-nodes -t ROUTE               # every route in the app
devlens blast-radius "src/auth/login.ts::login"
devlens reindex                          # index everything up front

Full reference: src/cli/README.md for every command with options and examples.

How to summarize

Summaries are per-node AI descriptions that make querying much richer. Run them during analysis, or any time later:

devlens analyze . --summarize                  # during analysis (recommended)
devlens summarize .                            # current directory's repo
devlens summarize <graphId>                    # a stored graph
devlens summarize . --force-summarize          # re-generate even if summarized
devlens summarize . --provider openai --provider-name deepseek --model deepseek-v4-flash

When a repo is re-summarized, only nodes without summaries are processed, so tokens are not spent twice. Configure your provider once with devlens init.

Configuration

Config lives in ~/.devlens/config.json, set via devlens init or devlens config. Models are discovered dynamically from each provider's /models endpoint, so there are no hardcoded model lists — pick whatever fits your budget and quality bar.

devlens config --set                            # interactive setup
devlens config --provider openai --provider-name deepseek --model deepseek-v4-flash --api-key <key>
devlens config --active openai:deepseek         # switch saved provider
devlens doctor                                  # health check

What DevLens understands

A graph is per repo and per language. Node types include FILE, FUNCTION, METHOD, CLASS, INTERFACE, STRUCT, ENUM, TRAIT, IMPL_BLOCK, ROUTE, COMPONENT, HOOK, STATE_STORE, UTILITY, TEST, STORY, and THIRD_PARTY (language dependent). Every node carries an importance score plus its functional, technical, and security summaries once summarized.

Edge types (the connections the graph draws): CALLS, IMPORTS, READS_FROM, WRITES_TO, PROP_PASS, EMITS, LISTENS, WRAPPED_BY, GUARDS, HANDLES, TESTS, USES, NEXTJS_API_CALL, NAVIGATES_TO, IMPLEMENTS (class to interface, trait, or ABC), EXTENDS (class to base class). EXPORTS, THROWS, MODULE, and PACKAGE are reserved for future languages.

Router awareness: routes are real graph nodes, for Next.js (app and pages), React Router, TanStack Router, wouter, Express, Fastify, Hono, Koa, Django URLconf and DRF, Flask blueprints, @RestController (Spring), Gin, Echo, chi, plain HTTP handlers, axum, actix, and rocket.

Search index

Each analyzed commit gets a derived BM25F search index next to the graph, at ~/.devlens/graphs/<graphId>/commits/<commitHash>.search.json.

• Every devlens analyze and devlens summarize rebuilds the index for the commit it produced, so search is ready immediately.
• Graphs built by older DevLens versions have no index. They are indexed lazily on the first search, or you can index everything up front with devlens reindex.
• The index is derived data: deleting it is safe, it rebuilds on demand.

devlens reindex                                  # latest commit of every graph
devlens reindex <graphId>                        # every commit of one graph
devlens reindex <graphId> <commitHash> --force   # one commit, rebuilt even if present

The same operations are available over HTTP: POST /api/reindex and POST /api/reindex/:graphId?commitHash=...&force=1.

Screenshots

Graph explorer · node inspector with summaries and security risk · focused subgraph · security findings

Why it's fast and cheaper

A node summary is roughly 50 tokens. The file it describes is roughly 2,000. Querying summaries and graph slices costs a fraction of reading files, so humans get answers faster and AI agents spend dramatically fewer tokens on the same task.

Who is this for

• Developers and teams: onboard in hours not weeks, review PRs with impact context, catch circular dependencies and god-files, keep living documentation.
• Engineering leaders: a bird's-eye architecture view, spot debt before it becomes a crisis.
• AI-augmented developers: stop letting your agent burn tokens re-reading files; it queries the graph instead.

Repository layout

devlensOSS/
├── src/
│   ├── cli/                  # `devlens` CLI (commander program + commands)
│   ├── core/                 # Shared query core (CLI + MCP, they never drift)
│   ├── mcp/                  # MCP server (stdio + HTTP) and packet renderer
│   ├── search/               # Local BM25F search stack (index, query, cache)
│   └── server/               # HTTP API for the Web UI + Next.js proxy
├── frontend/                 # Next.js graph visualizer (Cytoscape)
├── plugins/devlens/          # Deprecated agent skill source (old installs)
├── packages/skill-installer/ # @devlensio/skill (deprecated)
├── bin/ + npm/<platform>/    # Launcher and prebuilt binary packages
├── scripts/                  # Release tooling + start.mjs (bun start build-or-skip)
└── server.json               # MCP registry manifest

The analysis engine (native parsers and graph build) ships as the separate devlensio package.

DevLens Cloud

A hosted version is in development: shareable graphs, team workspaces, live per-commit documentation with node summaries, detailed PR analysis (summary, impact, security), commit diffs showing added/modified/deleted nodes, an AI-native chat over your codebase, and no local setup.

Join the waitlist →

License

AGPL-3.0. Part of the devlensio family of tools.