Pare
Provides MCP servers that wrap common developer tools (git, npm, docker, etc.) returning structured JSON output, enabling AI agents to reliably interact with these tools without parsing fragile terminal text.
README
<h1><img src="assets/logo.png" alt="" width="80" valign="middle" /> Pare</h1>
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:
| Tool Command | Raw Tokens | Pare Tokens | Reduction |
|---|---|---|---|
docker build (multi-stage, 11 steps) |
373 | 20 | 95% |
git log --stat (5 commits, verbose) |
4,992 | 382 | 92% |
npm install (487 packages, warnings) |
241 | 41 | 83% |
vitest run (28 tests, all pass) |
196 | 39 | 80% |
cargo build (2 errors, help text) |
436 | 138 | 68% |
pip install (9 packages, progress bars) |
288 | 101 | 65% |
cargo test (12 tests, 2 failures) |
351 | 190 | 46% |
npm audit (4 vulnerabilities) |
287 | 185 | 36% |
Token estimates use ~4 chars/token. The biggest savings appear on verbose commands (builds, installs, tests). For simpler tools like
eslintortsc, 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 itstructuredContent— 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 <file>..." to unstage)
modified: src/index.ts
new file: src/utils.ts
Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git restore <file>..." to discard changes in working directory)
modified: README.md
Untracked files:
(use "git add <file>..." 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.
| Category | Servers | Tools | Wraps |
|---|---|---|---|
| Version Control | git, github | 55 | git, gh |
| Languages & Packages | npm, python, cargo, go, deno, bun, nix, dotnet, ruby, swift, jvm | 101 | npm, pip, cargo, go, deno, bun, nix, dotnet, gem, swift, gradle, maven |
| Build, Lint & Test | build, lint, test, cmake, bazel | 23 | tsc, esbuild, vite, webpack, eslint, prettier, biome, vitest, pytest, jest |
| Infrastructure | docker, k8s, infra, security, remote | 40 | docker, kubectl, helm, terraform, ansible, trivy, ssh |
| Utilities | search, http, make, process, db | 21 | ripgrep, fd, curl, make, just, psql, mysql, redis, mongosh |
Tool Schemas — detailed response examples and field descriptions for every tool. See also Tool Response Examples 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
| Claude Code | Claude Desktop | Cursor |
| VS Code / Copilot | Windsurf | Cline / Roo Code |
| OpenAI Codex | Gemini CLI | Zed |
| Continue.dev |
Full Quickstart Guide — presets, ecosystem mapping, validation
Manual Configuration — config paths and formats for all clients
Agent Integration Guide — 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
| Env Var | Scope | Format | Example |
|---|---|---|---|
PARE_TOOLS |
All servers | server:tool,... |
git:status,npm:install |
PARE_{SERVER}_TOOLS |
One server | tool,... |
status,log,diff |
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"
}
}
}
}
Troubleshooting
| Issue | Solution |
|---|---|
npx not found / ENOENT on Windows |
Use cmd /c npx wrapper (see your client's setup guide) |
| Slow first start | Run npx -y @paretools/git once to cache, or install globally: npm i -g @paretools/git |
| Node.js version error | Pare requires Node.js >= 20 |
| NVM/fnm PATH issues | Use absolute path to npx: e.g., ~/.nvm/versions/node/v22/bin/npx |
| MCP connection timeout | Set MCP_TIMEOUT=30000 for Claude Code, or increase initTimeout in client config |
| Too many tools filling context | Use tool selection env vars to limit tools, or only install the servers you need |
Contributing
Each server is a self-contained package. See CONTRIBUTING.md for the full guide.
License
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。