See what your AI edits affect.
The 126× context-size result compares a simulated graph-assisted grep loop. It does not measure total agent-session token savings. Read the measured results and limits.
Built for the humans and the agents editing the code
codeweb computes one structural graph and serves it two ways, so you and the agents editing your code share the same map.
An interactive map
One self-contained HTML report — the dependency web, duplication, hotspots, and ranked, cycle-safe consolidation. No CDN, no network, opens straight from disk.
28 deterministic tools
Over MCP, your agents ask what already does this, what an edit breaks, and which fix is safe — before they write a line. No LLM in the analysis loop, so answers don't drift.
A four-stage deterministic pipeline
Parse-free extraction on the fast path for JavaScript, TypeScript, Python, Rust, Go, Java, C#, Ruby, PHP, Kotlin, Swift, C, and C++, with an agent fallback for everything else.
extract
Parse-free atomic nodes + edges, per language.
cluster
Group nodes into semantic domains by call gravity.
overlap
Body-confirm cross-domain duplication and rank it.
render
A self-contained interactive map + 28 agent tools.
Your repo — and any repo you're considering
Map your own codebase
See the real dependency web, find where logic is duplicated, and get an ordered, cycle-safe plan to consolidate it.
Review before you adopt
Clone a third-party repo read-only, map its supported source, and get an adoption review before you commit to depending on it. codeweb never runs the target.
Every claim comes with its data
Tested against independent checkers. The failures are published alongside the wins.
126×
smaller context size than a simulated graph-assisted grep loop. This is a context-size comparison, not total agent-session token savings.
0 disagreements
across ~490,000 checks of codeweb's answers against independent implementations: who calls what, what an edit breaks, where the cycles are.
44% → 74%
of a function's callers found by AI agents: grep vs codeweb, same context budget, better in all 5 runs (v0.9.0 pilot).
Approachable at the surface, deep when you need it
For a new codebase, open the map, follow the foundations-first reading order, and inspect the domains. For work at scale, add the CI gate and give your agents all 28 MCP tools. codeweb supports graphs with more than 16,000 symbols.
- Beginners — one command, one map, plain-language findings.
- Advanced — deterministic tools, fitness rules, sharded subgraphs, a PR gate.
The live report — a real map of axios (280 symbols, 7 domains). Open the interactive demo →
Direct answers to the questions people actually ask
The same answers your agents get — short, measured, and sourced. The numbers trace to the evidence ledger.
What is codeweb?
codeweb is an MCP server and Claude Code plugin that reads your code and maps it: every function, and every call between them (~3s for 3,000 symbols). Static analysis, no LLM — the same code always produces the same map. Coding agents query the map over 28 deterministic MCP tools; you get a self-contained interactive map. Free & MIT.
What breaks if I change this function?
The question codeweb exists to answer: codeweb_impact returns the blast radius — every function transitively affected by a change — before the edit is written. Measured on 30 vite symbols, one ~1KB codeweb call replaced a ~130KB grep loop: 126× cheaper.
Why do coding agents need a call graph instead of grep?
grep misses more than half of a function's real callers, and agents break the code they can't see. In measured tests, agents found 74% of a function's real callers with the map and 44% with grep, at the same context cost (v0.9.0 pilot, better in all 5 runs).
Does codeweb send my code anywhere or execute it?
No. codeweb runs entirely on your machine: no accounts, no telemetry, zero runtime dependencies — and it reads code, never executes it. No LLM in the analysis loop, so the same repo always yields the same graph.
Which languages does codeweb support?
13 languages natively — JavaScript, TypeScript, Python, Rust, Go, Java, C#, Ruby, PHP, Kotlin, Swift, C, and C++ — with an agent fallback for everything else.
How do I install codeweb?
Three ways: the Claude Code plugin (/plugin marketplace add GhostlyGawd/codeweb, then /plugin install codeweb), a one-shot map of the current repo (npx -y @ghostlygawd/codeweb .), or the MCP server for any MCP client (claude mcp add codeweb -- npx -y -p @ghostlygawd/codeweb codeweb-mcp). Node ≥ 22.
Ship bigger changes with fewer surprises
You and your agents see the same map before any of you edits. You find the right code faster, and you break less of it. Every claim on this site is measured, and the failures are published too.
To receive release notes, select Watch → Custom → Releases on GitHub. Release notes are the only announcement channel.