MCP Hub
Back to servers

Squire

CLI-first remote runtimes for validation and offload tasks, exposed over MCP stdio.

Registryglama
Stars
1
Updated
Apr 2, 2026
Validated
Apr 24, 2026
View Source
Claim this server

Squire CLI

Squire is a CLI for running short validation and offload jobs in clean remote runtimes. Use it for cross-environment checks, fresh validation loops, target compilation checks, ephemeral SQL sandboxes, and short heavy jobs that are awkward to run locally.

Quick start

curl -fsSL https://squire.run/install.sh | bash
squire login
squire --help

Defaults to:

  • https://api.squire.run

Works on:

  • macOS
  • Linux
  • WSL

Update later with:

squire update

Command discovery

Use these first:

  • squire --help shows the command catalog
  • squire --help --json shows the command catalog in machine-readable form
  • squire <command> --help shows command-specific usage

Prefer --json when another tool or agent will read the result.

For the full generated command catalog, public-policy notes, and examples, see:

  • docs/commands.generated.md

Task-to-command mapping

Use Squire when a task is environment-sensitive, target-sensitive, or short but heavy.

  • shell, Python, or Node runtime validation -> squire verify
  • short clean test runs -> squire test
  • lint or static analysis -> squire lint
  • Go or Rust target compilation -> squire compile
  • offline Qiskit Aer simulation -> squire quantum simulate
  • SQLite or Postgres validation -> squire sql
  • dependency, security, secret, or static checks -> squire audit
  • solver tasks -> squire solve
  • pandas, polars, or pyarrow jobs -> squire data
  • ffmpeg or media transforms -> squire media
  • offline headless browser verification -> squire browser
  • packaging or build sanity checks -> squire build
  • short comparative timing runs -> squire bench

Public-service note:

  • deps exists in the CLI, but is currently disabled on the public service under the zero-egress policy

Agent workflows

Squire is intentionally CLI-first and works well in terminal-first coding-agent workflows such as Claude Code, Codex, and shell-driven automation.

Copy and trim the example instruction files in docs/agent/ for your own environment:

  • docs/agent/CLAUDE.md.example
  • docs/agent/SKILLS.md.example
  • docs/agent/CODEX.md.example
  • docs/agent/squire-usage-guidelines.md
  • docs/agent/tool-constraints.md

Prefer local execution for tiny trivial checks. Prefer Squire when correctness depends on a fresh environment or when the task is too annoying to run locally.

For per-tool runtime, network, input, output, and public-service constraints, start with:

  • docs/agent/tool-constraints.md
  • docs/commands.generated.md

Current public-service limits

  • browser is offline-only and does not fetch remote http:// or https:// URLs
  • deps is currently disabled on the public service because sandbox egress is not allowed
  • audit supports secret scanning and local-config static analysis; dependency audit and remote Semgrep configs are disabled
  • quantum simulate is offline-only and currently requires trusted access or higher on the public service

Artifact downloads

Artifact-producing commands can download outputs locally:

squire media --script square.py --input image.png --download-artifacts ./out --json

Use --download-artifacts <dir> for:

  • squire data
  • squire media
  • squire browser
  • squire build
  • squire quantum simulate

Command examples

squire verify --lang bash --file script.sh --targets alpine-3.20,ubuntu-24.04,debian-12 --json
squire test --lang python --file test_app.py --cmd "pytest -q" --targets py310,py311 --json
squire lint --lang python --tool ruff --file app.py --json
squire quantum simulate --file shor.py --shots 2048 --download-artifacts ./quantum-out --json
squire audit --secrets --path src --json
squire build --lang python --file pyproject.toml --path src --targets manylinux,musllinux --download-artifacts ./dist --json
squire bench --lang python --file bench.py --targets py310,py311 --json
squire browser --path website/public --screenshot page.png --download-artifacts ./browser-out --json
squire sql --dialect sqlite --query "SELECT 1" --json
squire compile --lang go --file main.go --targets linux/amd64,linux/arm64 --json
squire solve --solver z3 --file constraints.smt2 --json
squire data --script transform.py --input big.csv --download-artifacts ./data-out --json
squire media --script clip.py --input image.png --download-artifacts ./media-out --json

A tiny runnable quantum example is in examples/quantum/shor.py.

Optional MCP mode

Bootstrap MCP auth with:

squire mcp login

This logs you in and prints the SQUIRE_TOKEN and SQUIRE_API_BASE_URL values an MCP host needs.

Then either run the local stdio wrapper:

squire mcp serve

This exposes the same Squire command surface over MCP stdio and reuses the current local Squire session. Squire remains CLI-first; MCP is an optional wrapper.

For Cline on a machine that already has a working local Squire session, the shortest path is:

cline mcp add squire -- squire mcp serve

Squire is also published to the MCP Registry as io.github.reidgoodbar/squire. The registry package is a GHCR-backed OCI stdio server that runs the same squire mcp serve surface and expects:

  • SQUIRE_TOKEN required
  • SQUIRE_API_BASE_URL optional

Registry metadata lives in registry/server.json.

For Cline and other agent-hosted setup, see:

  • llms-install.md

Local build

go build -o ./bin/squire ./cmd/squire

Release packaging

./scripts/build-release.sh

Artifacts go to dist/ and use the same stable names that https://squire.run/install.sh downloads.

Reviews

No reviews yet

Sign in to write a review