GitHub

Pare

Reliable, structured CLI output for AI agents — no more parsing fragile terminal text.

Pare provides MCP servers that wrap common developer tools (git, npm, docker, test runners, etc.) and return clean, schema-validated JSON instead of raw terminal text. Agents get typed data they can act on directly, without brittle string parsing.

The Problem

Parsing CLI output is fragile. Raw terminal text includes ANSI escape codes, decorative headers, progress bars, locale-specific formatting, and platform differences that break agent workflows in subtle ways. An agent that works fine with git status on macOS may fail on Windows because the output format changed. A test runner's summary line might shift between versions, silently breaking a regex.

Pare eliminates this entire class of errors by returning schema-validated JSON with consistent field names, regardless of platform, tool version, or locale. As a bonus, structured output is significantly smaller — agents use fewer tokens per tool call:

Token estimates use ~4 chars/token. The biggest savings appear on verbose commands (builds, installs, tests). For simpler tools like eslint or tsc, the main advantage is reliable structured data — agents can use typed JSON directly rather than parsing strings.

How It Works

Each Pare tool returns two outputs:

• `content` — human-readable text, for MCP clients that display it
• `structuredContent` — typed, schema-validated JSON, ready for agents to process

This uses MCP's structuredContent and outputSchema features to provide type-safe, validated data that agents can rely on without custom parsing.

Example: git status

Raw git output (~118 tokens):

On branch main

Your branch is ahead of 'origin/main' by 2 commits. (use "git push" to publish your local commits)

Changes to be committed: (use "git restore --staged ..." to unstage) modified: src/index.ts new file: src/utils.ts

Changes not staged for commit: (use "git add ..." to update what will be committed) (use "git restore ..." to discard changes in working directory) modified: README.md

Untracked files: (use "git add ..." to include in what will be committed) temp.log

Pare structured output (~59 tokens):

{

"branch": "main", "upstream": "origin/main", "ahead": 2, "staged": [ { "file": "src/index.ts", "status": "modified" }, { "file": "src/utils.ts", "status": "added" } ], "modified": ["README.md"], "deleted": [], "untracked": ["temp.log"], "conflicts": [], "clean": false }

50% fewer tokens. Zero information lost. Fully typed. Savings scale with output verbosity — test runners and build logs see 80–92% reduction.

Available Servers (28 packages, 240 tools)

Install only the servers relevant to your stack — most projects need just 2–4. The full catalog covers a wide range of ecosystems so Pare works wherever you do.

[Tool Schemas](./docs/tool-schemas/) — detailed response examples and field descriptions for every tool.

See also [Tool Response Examples](./docs/tool-response-examples.md) for quick JSON samples.

Quick Setup

# 1. Configure MCP servers (non-interactive)

npx @paretools/init --client claude-code --preset web

2. Add agent rules to your project

(append to existing CLAUDE.md, or copy if new)

cat node_modules/@paretools/init/rules/CLAUDE.md >> CLAUDE.md

3. Restart your client session

4. Validate

npx @paretools/init doctor

Available presets: web, python, rust, go, jvm, dotnet, ruby, swift, mobile, devops, full

Setup Guides by Client

[Full Quickstart Guide](./docs/quickstart.md) — presets, ecosystem mapping, validation

[Manual Configuration](./docs/manual-configuration.md) — config paths and formats for all clients

[Agent Integration Guide](./docs/agent-integration.md) — rule files, hooks, CLI-to-MCP mapping

Configuration

Tool Selection

By default, every Pare server registers all of its tools. If a server exposes tools you don't need — or you want to limit which tools are available to an agent — you can filter them with environment variables.

Per-server filter — restrict a single server's tools:

# Only register status and log in the git server

PARE_GIT_TOOLS=status,log npx @paretools/git

Universal filter — restrict tools across all servers:

# Only register these specific tools across any server

PARE_TOOLS=git:status,git:log,npm:install npx @paretools/git

Disable all tools — set the env var to an empty string:

PARE_GIT_TOOLS= npx @paretools/git   # no tools registered

Rules:

• No env var = all tools enabled (default)
• PARE_TOOLS (universal) takes precedence over per-server vars
• Server names use uppercase with hyphens replaced by underscores (e.g., PARE_MY_SERVER_TOOLS)
• Whitespace around commas is ignored

Common patterns:

# Read-only git (no push, commit, add, checkout)

PARE_GIT_TOOLS=status,log,diff,branch,show

Minimal npm

PARE_NPM_TOOLS=install,test,run

Only specific tools across all servers

PARE_TOOLS=git:status,git:diff,npm:install,test:run

In JSON MCP config, set via the env key:

{

"mcpServers": { "pare-git": { "command": "npx", "args": ["-y", "@paretools/git"], "env": { "PARE_GIT_TOOLS": "status,log,diff,show" } }