
dep-scout
io.github.unievolverPublished by GitHub account unievolverv0.2.0Updated Oct 25, 2026
Help AI agents find and evaluate mature packages across 6 ecosystems instead of reinventing them.
Overview
A local MCP server that lets a coding assistant search six package registries for mature libraries, score their reuse quality, and check security and license.
- What you can do
- Search crates.io, npm, PyPI, Go, Maven Central and NuGet for packages matching a feature description, returning ranked candidates with a 0–100 reuse score and warnings. Inspect a single package by exact name and ecosystem for license, known OSV vulnerabilities, deprecation or yank status, version requirements, popularity and last-update recency. Search the official MCP registry for existing MCP servers so the assistant reuses one instead of building it.
- When to use it
- Useful when an AI coding agent is about to write a feature from scratch and should first check whether a maintained package already solves it, or when you want a quick read on a dependency's health, license and known vulnerabilities before adopting it.
- Requirements
- Runs as a local stdio process, distributed as an .mcpb bundle or built from source with a recent Rust toolchain (edition 2024). No API keys or environment variables; all data sources are public keyless APIs, so network access is needed.
Installation
In SourceWeft
- Open dep-scout in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
dep-scout
先查再造 (research before you build) — an MCP server that stops AI coding agents from reinventing the wheel.
Before an agent writes a feature from scratch, dep-scout searches the right registry — crates.io (Rust), npm (JS/TS/frontend), PyPI (Python), pkg.go.dev (Go), Maven Central (Java/Kotlin) or NuGet (.NET) — for a mature package that already solves the problem, and scores each candidate on reuse quality (popularity, maintenance recency, version stability, metadata) — and flags known security advisories (OSV) and license-compliance risks. The agent then reuses a proven, safe solution instead of hand-rolling buggy code. It can also search the official MCP registry so the agent reuses an existing MCP server instead of building one.
Works with Cursor (and any MCP client) over stdio. No API keys — every data source is a keyless public API.
Why
AI now writes a lot of code — and its worst habit is reimplementing solved problems, picking abandoned libraries, or hallucinating packages. dep-scout turns the best practice "don't build it if a maintained package already does it" into a tool the agent calls automatically, across languages.
Tools
ecosystem accepts aliases: js/ts/javascript/typescript/node/frontend → npm,
py/pip → python, golang → go, java/kotlin/gradle/jvm → maven,
dotnet/csharp/net → nuget, rs/cargo → rust.
Reuse score (0–100)
- Popularity (≤40) — downloads normalised to a ~90-day window; for Go/Maven (no public download stats) GitHub stars are used as a proxy.
- Maintenance (≤30) — how recently the package was updated.
- Stability (≤15) — stable
1.0+vs0.x; penalty for pre-releases. - Metadata (≤15) — repository, docs/homepage, license present.
- Security penalty — known OSV advisories on the resolved version sink the score and de-rank the candidate.
- License flag — copyleft / commercially-restricted licenses (GPL, AGPL, LGPL, MPL, SSPL, BUSL…) are surfaced as compliance warnings.
- Penalty — latest version yanked (crates/PyPI) or deprecated (npm).
Verdicts: ≥78 ✅ strongly recommend · 58–77 🟡 usable · 38–57 🟠 cautious ·
<38 🔴 find an alternative.
Ranking blends textual relevance (query-term overlap, exact-name match) with the reuse score and the registry's own ordering, so results are both on-topic and high quality.
Data sources (all keyless)
Python note: PyPI has no public full-text search API (the legacy XML-RPC search is gone and the web search is bot-blocked), so Python search is name-oriented via deps.dev. It excels when the query is/contains the library name; for pure capability phrases, prefer
inspect_packagewith a known name.
Build
Requires a recent Rust toolchain (edition 2024).
The binary is produced at target/release/dep-scout (.exe on Windows).
Use in Cursor
-
Register the MCP server. A portable project config lives at
.cursor/mcp.json— it usescargo run --releaseso no machine-specific paths are required. For global use, add an entry to~/.cursor/mcp.json:On Windows use
...\\target\\release\\dep-scout.exe. Build the release binary first (cargo build --releaseabove) when using a direct binary path. -
Restart Cursor; you should see
dep-scoutwith its tools enabled. -
The
.cursor/rules/research-before-build.mdcrule makes the agent search before building. A portable, client-agnostic version lives atskill/SKILL.md.
How it talks
Standard MCP over stdio (JSON-RPC 2.0), built on the official
rmcp SDK. Logs go to stderr so they never
corrupt the protocol stream on stdout.
Roadmap
- More ecosystems: Ruby (RubyGems), PHP (Packagist), Dart/Flutter (pub.dev).
- Per-ecosystem popularity calibration (npm/PyPI volumes dwarf crates.io).
- Last-update recency for NuGet; richer Maven search ranking.
- Deeper repo health signals (open issues, release cadence, maintainer count).
- A curated trust/quality dataset — the real moat over a plain registry mirror.
Project layout
MCP Registry
Published to the official MCP Registry as
io.github.unievolver/dep-scout. Download the .mcpb bundle from
GitHub Releases or build from
source with cargo build --release.
- MCP Registry name:
mcp-name: io.github.unievolver/dep-scout
License
MIT
Source: README.md at commit 0de4f9e
Tools
0Version history
1- v0.2.0LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts