Pare

Pare is a collection of MCP servers that wrap common developer tools (such as git, npm, docker, and testing frameworks) and return clean, schema-validated JSON outputs instead of raw terminal text. This makes tool output easier for AI agents to consume and reason about, reducing token usage and parsing errors. The Pare server set provides two outputs per tool call: content (human-readable) and structuredContent (typed, schema-validated JSON). You can invoke various wrapped tools through the Pare MCP endpoints and rely on the structuredContent to drive automated workflows, conditionals, and decision making in agent-based scenarios.

To start using Pare, deploy the MCP server via your preferred package manager (for example via npx as shown in the config). Once running, request tool actions such as git status, npm install, docker build, or test runners (pytest, jest, vitest, etc.). The responses will include a structuredContent payload with clearly typed fields, statuses, and results, enabling downstream agents to interpret results without needing to parse human-friendly CLI output. If you need multiple tools, Pare exposes a variety of servers (git, npm, docker, test, etc.) that can be combined in your MCP workflow.

Prerequisites:

• Node.js >= 20
• npm or corepack

Installation steps:

1. Install Node.js (if not already installed):

   • On macOS: brew install node
   • On Windows: download and run the Node.js installer from https://nodejs.org/
   • On Linux: use your distro’s package manager or nvm to install Node.js 20+

2. Install the Pare MCP server package (example using npm via npx): npm install -g @paretools/git OR npx -y @paretools/git

3. Run the Pare MCP server (configured to wrap git and related tools): npm run pare-server OR npx -y @paretools/git

4. Verify the server is responding (example): curl http://localhost:PORT/health

Notes:

• If you are using a different tool within Pare (e.g., npm, docker, test), substitute the corresponding package name in the install/run commands.
• Ensure network access and any required credentials (see environment variables) are configured if you’re accessing private registries or services.

Tips and caveats:

• Pare returns two outputs for each tool call: content (human-friendly) and structuredContent (validated JSON). Use structuredContent for automation.
• If you plan to run multiple Pare tools in a single workflow, consider batching requests to minimize latency.
• Common environment variables to consider adding:
  • PARE_API_TOKEN or PARE_API_KEY for authenticated tool access
  • PARE_LOG_LEVEL to control verbosity (e.g., info, warn, error)
  • PARE_TIMEOUT_MS to cap operation timeouts
• If parsing errors occur, inspect the raw content output in addition to structuredContent to diagnose schema mismatches or tool-specific issues.
• Pare supports a broad range of tools (git, github, search, test, npm, docker, build, lint, http, etc.). Check the repository’s available server packages to tailor your MCP workflows.