ContextKit is a CLI tool and MCP client used to create spec files with embedded context for AI coding agents. It generates specs from reusable Jinja2 templates and populates context from various MCP sources and user input. The official description specifically notes that although ContextKit is an MCP client, it does not use AI models; instead, it is a tool that leverages the MCP ecosystem for context injection.
In terms of purpose, ContextKit uses the MCP (Model Context Protocol) ecosystem to inject context from sources such as ticketing systems, databases, document stores, and design tools into spec files. It works around "spec templates": these templates are reusable Jinja2 template files that contain context variables and are used to define the general structure and requirements of development tasks. Context variables are automatically obtained from MCP resources when possible, and when additional input is needed, they are collected interactively from the user.
Main capabilities include: a Jinja2-based template engine, with additional global functions call_tool(server_name, tool_name, arguments) (calls an MCP tool) and get_resource(server_name, resource_uri) (retrieves an MCP resource), while also supporting all standard Jinja2 filters and functions. Template variables support direct values (such as --var ticket_id=ACME-123), JSON objects (such as --var ticket='{"id": 123, "status": "open"}'), and when not provided, prompt for input interactively. It also supports filtering and processing context (such as filtering sensitive information and extracting specific fields), as well as passing template content through a pipe.
In terms of usage, the environment requirement is Python 3.11 or higher. It can be installed from source and also supports installation in editable mode; after editable installation, the cxk command can be used directly, or it can be used without editable installation. The typical workflow is: 1. Initialize a project in the project directory (this creates a .cxk/ directory for storing configuration); 2. Add an MCP server; 3. Create a spec from a template. CLI commands are divided into three categories: project management, MCP server management, and spec generation. Templates can contain variables, for example, variables in the template spec.md will prompt the userdirectly provide the value of prd, or choose an MCP tool to call; templates with MCP tool calls and templates with MCP resources are also supported.
The applicable scenarios are: development teams that need to prepare spec files with ready-made context for AI coding agents and want to reuse spec templates in version control and keep team spec formats consistent. The officially listed values include: fast (automatically prefill specs with relevant context), lower cost (AI agents do not need to make additional MCP calls), clean codebase (keep reusable spec templates in version control), consistency (standardized spec formats across teams), extensible (leverage the growing MCP ecosystem).
In terms of limitations, the official documentation clearly states that ContextKit does not use AI models; it is a tool for context injection, not an AI generation tool. It requires Python 3.11 or higher. Regarding specific charges, pricing, platform support scope, and more detailed usage steps, the current materials do not yet provide them.
