vdiff
Breaking-change diffs for npm packages, served over REST and MCP, so coding agents stop writing code against outdated API knowledge.
README
vdiff
A breaking-change diff API for npm packages. Given a package and two versions, it returns a structured, machine-readable list of what actually broke: removed exports, changed function signatures, and removed or changed class and interface members. Each entry includes the before and after signatures plus a short migration note.
Live at https://vdiff-api.onrender.com. Most easily consumed through the MCP server, vdiff-mcp:
claude mcp add vdiff -- npx -y vdiff-mcp
Or call the REST API directly:
curl "https://vdiff-api.onrender.com/v1/diff?ecosystem=npm&package=zod&from=3.24.0&to=4.0.0"
Why this exists
LLMs learn a package's API surface during training, then the package moves on. When a coding agent writes code against zod or express, it writes for the version it remembers, which is often not the version in your lockfile. The result is confident code that calls functions that were renamed or removed two majors ago.
Existing tools only solve part of this. Version lookup tools tell the agent what the current version is. Documentation tools tell it what the docs say today. Neither answers the question the agent actually has mid-edit: "what changed between the version I know and the version installed here?"
vdiff answers exactly that. Diffs are computed from the package's own type declarations rather than changelogs, so the output reflects the real exported surface, and every response carries a confidence score so the consumer knows how much to trust it.
Endpoint reference: docs/api.md.
How it works
- Resolve. Fetch package metadata, versions and dist-tags from the npm registry.
- Extract. For each of the two versions, download the tarball, keep only the
.d.tsfiles, and build a table of public exports (functions, classes, members, normalized call signatures) using the TypeScript compiler API. Bundled declarations are preferred; packages that ship none fall back to the matching DefinitelyTyped@types/*package, version-matched bymajor.minor. - Compare. Diff the two export tables into typed change entries (
export_removed,signature_changed,member_removedand so on), each with before/after signatures and a migration note. - Cache and meter. Results are stored in Postgres keyed on (package, from, to), so each version pair is computed once, ever. Every request is logged with cache-hit status and user agent.
- Guard. Per-IP rate limits, a cap on simultaneous diff computations, size limits on tarballs and extracted declarations, and fetch timeouts keep the service safe to expose publicly.
Diffs are type-level: a runtime behavior change that leaves the types untouched is invisible. Responses using bundled types carry confidence 0.9; responses using community-maintained @types/* declarations carry 0.8.
API overview
| Endpoint | Purpose |
|---|---|
GET /v1/diff |
Breaking-change diff between two versions (from required, to defaults to latest) |
GET /v1/resolve |
Latest version and dist-tags for a package |
GET /healthz |
Liveness check |
See docs/api.md for parameters, response shapes, change types, error codes and rate limits.
Stack
| Layer | Choice | Why |
|---|---|---|
| Runtime | Node 20+, TypeScript | npm-only .d.ts diffing needs the TS compiler API, so one language |
| API | Fastify 5 | Fast, minimal, good TypeScript support |
| Diffing | typescript compiler API |
Structured symbol tables from .d.ts files, the core of the product |
| Database | Postgres 18 (Docker local, Neon prod) | JSONB for variable-shape diff payloads, SQL for billing and analytics |
| Registry | npm registry HTTP API + tar |
Packuments and tarball extraction, declarations only |
| Tests | Vitest | Unit tests for compare logic and @types version matching |
The code is cloud-agnostic: a plain container plus a DATABASE_URL. It currently runs on Render with Neon Postgres.
Running it yourself
docker compose up -d # Postgres 18 on :5432
npm install
npm run db:migrate # apply src/db/schema.sql
npm run dev # API on :3000
Or containerized, the way a PaaS runs it (applies the schema on boot, then serves):
docker build -t vdiff-api .
docker run -p 3000:3000 -e DATABASE_URL="postgres://user:pass@host:5432/db" vdiff-api
Configuration
All configuration is via environment variables:
| Env var | Default | Purpose |
|---|---|---|
PORT |
3000 |
Listen port |
DATABASE_URL |
local Docker Postgres | Postgres connection string |
RATE_LIMIT_DIFF_MAX |
30 |
Per-IP /v1/diff requests per minute |
RATE_LIMIT_RESOLVE_MAX |
120 |
Per-IP /v1/resolve requests per minute |
COMPUTE_CONCURRENCY |
2 |
Max simultaneous diff computations |
TRUST_PROXY |
unset | Set true behind a PaaS proxy so the rate limiter sees real client IPs |
Hardening notes
- Rate limiting: per-IP, in-memory (fine while single-instance).
/healthzis exempt for platform health checks. - Compute cap: at most
COMPUTE_CONCURRENCYuncached diffs compile at once; excess requests get a503withretry-after. Cached diffs are always served. - Size guards: tarball downloads are capped at 50 MB (checked via content-length and counted bytes), extracted declarations at 15 MB per version. Oversized packages fail with a clear
422. - Fetch budgets: 10 s for packuments, 30 s for tarballs. Tarballs are only fetched from
registry.npmjs.orgover HTTPS.
Project layout
src/
index.ts Fastify bootstrap, /healthz
routes.ts /v1/resolve, /v1/diff: validation, cache, dedup, metering
registry/npm.ts packument fetch, tarball download, .d.ts extraction
diff/
symbols.ts .d.ts to symbol table (TS compiler API)
compare.ts symbol table diff to breaking changes
engine.ts orchestration, @types/* fallback, confidence
db/
schema.sql packages, versions, diffs, diff_requests_log
migrate.ts applies schema
mcp/ vdiff-mcp, the MCP server wrapping this API (published to npm)
docs/
api.md endpoint reference, kept current with the code
Tests
npm test # unit tests (Vitest)
npx tsc --noEmit # typecheck
License
The API and diff engine are licensed under the Functional Source License, v1.1, MIT Future License (FSL-1.1-MIT): free to use, read and modify for anything except offering a competing service, and each release automatically becomes MIT two years after publication.
The MCP server wrapper in mcp/ is MIT licensed.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.