A CLI tool and MCP client for creating spec files for AI coding agents. ContextKit generates specs from reusable Jinja2-based templates, filling in context from various MCP sources and user input.

Note: While ContextKit is an MCP client, it doesn't use an AI model - it's a tool to harness the MCP ecosystem for context injection.

ContextKit utilizes the MCP (Model Context Protocol) ecosystem to inject context into spec files from various sources like ticketing systems, databases, document storage, and design tools. It works with spec templates - reusable Jinja2 template files containing context variables that define the common structure and requirements of development tasks.

Context variables are automatically fetched from MCP resources when possible, or collected interactively from the user when additional input is needed.

• Python 3.11 or higher

git clone <repository-url>
cd context-kit
uv sync

pip3 install -e .

After editable installation, you can use the cxk command directly:

cxk --help

Alternatively, without editable installation:

python cxk.py --help
# or with uv:
uv run cxk.py --help

Run this in your project directory. It creates a .cxk/ directory to store config.

cxk init

# Add SSE server
cxk mcp add-sse ticket-server http://localhost:3000

# Add stdio server with environment
cxk mcp add-stdio local-tools --env API_KEY=secret -- python server.py

# Add HTTP server
cxk mcp add-http doc-service http://localhost:8000

cxk create-spec path/to/template.md

Template (spec.md):

# Development Task
Please develop the task below. You should start by ...

## PRD
{{ prd }}

Generate spec:

cxk create-spec spec.md

This will prompt you to provide a value for prd either directly or by selecting an MCP tool to call.

Template (spec.md):

# Development Task
Please develop the task below. You should start by ...

{% set ticket = call_tool('ticket-service', 'get_issue', {'cloudId': '1234', 'issueKey': ticket_id}) %}

## Task Description
{{ ticket.fields.description }}

## Acceptance Criteria
{{ ticket.fields.acceptance_criteria }}

Generate spec:

cxk create-spec spec.md --var ticket_id=ACME-123

Template (design-spec.md):

## Design Requirements
{{ get_resource('figma-service', 'designs://'+design_id) }}

Generate spec:

cxk create-spec design-spec.md --var design_id=fig-456

Filter sensitive information:

{% set support_ticket = call_tool('support', 'getTicket', ticket_id) %}
## Support Request
{{ support_ticket | regex_replace(r'\b[\w.+-]+@[\w.-]+\.\w+\b', '[EMAIL_REDACTED]') }}

Extract specific fields:

{% set ticket = call_tool('ticket-service', 'get_issue', {'issueKey': ticket_id}) %}
## Summary
{{ ticket.fields.summary }}

## Priority
{{ ticket.fields.priority.name }}

When variables aren't provided via --var, ContextKit prompts interactively:

cxk create-spec template.md
# Prompts:
# ? How would you like to provide the value for 'ticket_id'?
#   › Provide value directly
#     Use MCP tool to fetch value
#     Read from file

cxk create-spec template.md --output result.md --var ticket_id=ACME-123

cat template.md | cxk create-spec --var ticket_id=ACME-123

# Initialize new project (creates .cxk/ directory)
cxk init

# Add SSE MCP server
cxk mcp add-sse <server-name> <websocket-url>

# Add stdio MCP server
cxk mcp add-stdio <server-name> [--env KEY=value] -- <command> [args...]

# Add HTTP MCP server  
cxk mcp add-http <server-name> <http-url> [--header KEY=value]

# Generate spec from template
cxk create-spec <template-path> [--var KEY=value] [--output <file>]

# Generate with multiple variables
cxk create-spec template.md --var ticket_id=ACME-123 --var env=production

ContextKit uses Jinja2 as its template engine with additional global functions:

• call_tool(server_name, tool_name, arguments) - Call MCP tool
• get_resource(server_name, resource_uri) - Get MCP resource
• All standard Jinja2 filters and functions

Variables can be:

• Direct values: --var ticket_id=ACME-123
• JSON objects: --var ticket='{"id": 123, "status": "open"}'
• Interactive input: Prompted when not provided

# Development setup
cxk init
cxk mcp add-sse jira https://mcp.example.com/jira
cxk mcp add-stdio local-db --env DB_URL=postgresql://... -- python db_server.py
cxk mcp add-http docs http://localhost:8000 --header Authorization='Bearer token'

• 🚀 Speed: Pre-populate specs with relevant context automatically
• 💰 Cost reduction: AI agents don't need to make additional MCP calls
• 🧹 Clean codebase: Keep reusable spec templates in version control
• 🔄 Consistency: Standardized spec format across your team
• 🔌 Extensible: Leverage the growing MCP ecosystem

# Install development dependencies
uv sync

# Run tests
uv run pytest

# Run linting
uv run ruff check
uv run ruff format