AlgoFox MCP — coding interview practice in your terminal

Practice coding interviews with the coding agent you already use — Claude Code, Cursor, Codex, or any MCP client.

Works fully offline. Nothing ever leaves your machine. No account, no telemetry, no network calls — problems ship in the package and your progress lives in ~/.algofox/.

How it works

1. Say "Continue AlgoFox" (or pick a problem). The agent materializes it into ./algofox/<slug>/: the problem statement, a starter file, and a test runner.

2. You write the solution — the agent is your interviewer/coach, not your ghostwriter. It clarifies, runs the tests when you ask, and fetches tiered hints.

3. Tests run locally (python3 run_tests.py or node run_tests.mjs). Results are recorded; solving earns XP and keeps your streak alive.

Close your terminal any time. Come back in any supported agent, say "Continue AlgoFox", and you resume where you left off.

Related MCP server: devcoach

Install

Claude Code (recommended: the plugin)

The algofox-skill plugin bundles this server plus an interview-coach skill and /algofox commands:

/plugin marketplace add phuongnd11/algofox-skill

Or add just the server:

claude mcp add algofox -- npx -y -p @algofoxdotapp/mcp algofox-mcp

Cursor

.cursor/mcp.json:

{ "mcpServers": { "algofox": { "command": "npx", "args": ["-y", "-p", "@algofoxdotapp/mcp", "algofox-mcp"] } } }

Codex

~/.codex/config.toml:

[mcp_servers.algofox]
command = "npx"
args = ["-y", "-p", "@algofoxdotapp/mcp", "algofox-mcp"]

For clients without skills, copy the coach block from AGENTS.md into your project's AGENTS.md / rules file so the agent behaves like an interviewer instead of solving problems for you.

Standalone CLI

Progress without opening an agent:

npx -p @algofoxdotapp/mcp algofox status     # resume card
npx -p @algofoxdotapp/mcp algofox progress   # full problem list
npx -p @algofoxdotapp/mcp algofox history    # recent activity
npx -p @algofoxdotapp/mcp algofox resume     # ready-to-paste agent prompt

What's inside

• 10 original coding problems (arrays & hashing to start — more topics coming) with typed test cases, tiered hints, and reference solutions.

• 5 languages: Python, JavaScript, TypeScript, Java, Go. You pick on first run; switch any time by telling your agent. LeetCode-style Solution boilerplate, sample input/output, and a test harness are generated from each problem's typed signature.

• Solutions are earned: reveal_solution unlocks after 2 failed runs or 2 hints. All content is local, so you can peek at the answer keys — but you'd only be cheating yourself.

• A run token ties submitted results to an actual test run, so your agent can't hand-wave a pass.

Want a full curriculum — guided lessons, spaced review, quick-play drills, company tracks? That's the AlgoFox app.

Requirements

Node ≥ 18. Per language: Python needs python3; TypeScript needs Node ≥ 22.6; Java needs a JDK (javac); Go needs the Go toolchain. JavaScript needs nothing extra.

License

Code: MIT. Problem content (content/): CC BY-NC-ND 4.0 — see content/LICENSE.