
Security News
Anthropic Identifies Biased Reasoning and Recklessness as Drivers of Claude’s PyPI Attack
Anthropic found biased reasoning and recklessness drove Claude Mythos 5 to publish malware on PyPI and compromise a security vendor.
A testing workbench for MCP (Model Context Protocol) servers — inspect, call, budget, and bench MCP servers from the command line.
A testing workbench for MCP (Model Context Protocol) servers. Inspect a server's tool surface, call tools, price the context-token budget of a tool set, and bench a real model against it — all from the command line.
This npm package is a thin installer/launcher: on install it downloads the right
prebuilt jig binary for your platform from
GitHub Releases, verifies its
SHA-256 checksum, and puts it on your PATH. The heavy lifting is done by the
native Rust binary — see the main repository
for full documentation.
No install needed — run it straight with npx the same way you run MCP servers:
npx @shodh/jig inspect --stdio "npx -y @modelcontextprotocol/server-everything"
Or install it globally:
npm install -g @shodh/jig
jig inspect --stdio "npx -y some-mcp-server"
Common commands:
jig inspect --stdio "<server cmd>" # handshake + list tools/resources/prompts
jig call --stdio "<server cmd>" --tool <name> --args '{"...":"..."}'
jig budget --stdio "<server cmd>" --model gpt-4o --model claude-sonnet
jig bench --stdio "<server cmd>" --task "book a table for two"
jig also speaks remote MCP over Streamable HTTP via --http <url> (with
repeatable --header "Authorization: Bearer …"). Run jig --help for the full
surface.
Exit codes are meaningful and forwarded faithfully by the launcher:
0 success · 1 a jig-level failure (transport/protocol/usage) · 2 a tool
call that the server reported as isError: true.
Prebuilt binaries are published for:
| OS | Arch | Node platform/arch | Release target |
|---|---|---|---|
| Windows | x64 | win32/x64 | x86_64-pc-windows-msvc |
| macOS (Intel) | x64 | darwin/x64 | x86_64-apple-darwin |
| macOS (Apple) | arm64 | darwin/arm64 | aarch64-apple-darwin |
| Linux | x64 | linux/x64 | x86_64-unknown-linux-musl |
On any other platform/arch the installer fails with a clear message pointing you
at the cargo build path:
cargo install --git https://github.com/Shodh-Labs/jig jig-cli
Every download is verified before it is trusted. The installer fetches the
release's SHA256SUMS alongside the archive and computes the SHA-256 of the
downloaded archive in memory; if it does not match the published digest, the
install aborts and nothing is extracted or written. On Windows the archive
is parsed entirely in-process (no shelling out), so the verified bytes never
touch disk before being trusted.
Publishing to npm uses npm provenance, so you can trace the published package back to the exact GitHub Actions run and commit that produced it.
| Variable | Effect |
|---|---|
JIG_BINARY_PATH | Use this local binary verbatim and skip all network access. For CI, air-gapped, and testing. |
JIG_DOWNLOAD_BASE | Download from this base URL instead of GitHub Releases. Layout: {base}/v{version}/{asset}. |
Examples:
# Use a binary you built or vendored yourself; no download happens.
JIG_BINARY_PATH=/path/to/jig npm install -g @shodh/jig
# Pull binaries from an internal mirror that mirrors the GitHub Releases layout.
JIG_DOWNLOAD_BASE=https://mirror.internal.example/jig npm install -g @shodh/jig
Node's built-in HTTP client does not honor HTTPS_PROXY/HTTP_PROXY, and
this installer keeps a zero-dependency footprint, so it connects directly.
If a proxy is configured the installer warns and continues; if the direct
connection then fails, download the release archive manually and point
JIG_BINARY_PATH at the extracted binary, or install via cargo (above).
.zip; the single binary entry is extracted with a
small in-process zip reader (Node has zlib but no zip reader) — no external
tools, no temp files..tar.gz; the binary is streamed out with the
system tar (always present on the supported unix targets). There is no
pure-JS tar fallback by design.Apache-2.0. See the main repository for source, issues, and full documentation.
FAQs
A testing workbench for MCP (Model Context Protocol) servers — inspect, call, budget, and bench MCP servers from the command line.
We found that @shodh/jig demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.

Security News
Anthropic found biased reasoning and recklessness drove Claude Mythos 5 to publish malware on PyPI and compromise a security vendor.

Research
/Security News
Malicious Chrome and Firefox extensions target Axiom Trade and Padre users, stealing session tokens and wallet data.

Security News
GPT-6 Astra hits 100% on ExploitBench and finds zero-days autonomously, while independent tests reveal scope violations and monitoring gaps.