MCP Server
Model Context Protocol server that exposes build tools to Claude Code. Provides 50+ MCP tools for builds, patterns, skills, and delegation.
More than just an MCP server - a complete autonomous development platform
Model Context Protocol server that exposes build tools to Claude Code. Provides 50+ MCP tools for builds, patterns, skills, and delegation.
Background service managing build queues, resource limits, and process monitoring. Task queue with persistent storage, watchdog enforcement.
Terminal interface for monitoring builds and managing the system. Mission Control TUI with real-time phase progress and multi-build support.
Everything you need for autonomous software development
Scout, Architect, Builder, Test, Docs, Deploy, Feedback - each phase spawns fresh Claude instances with isolated context.
Automatically fixes test failures through redesign/rebuild cycles. Up to 3 retry attempts with intelligent debugging.
AI decides when to spawn parallel agents for faster builds. 2-8 concurrent workers, 3x faster than sequential.
Smart change detection rebuilds only what changed. Caches scout analysis and test impact analysis.
SQLite database tracking issues, solutions, and build metrics. Full-text search and confidence scoring.
Reusable code implementations with success rate tracking. Search by query, filter by project type.
Post-phase validators, checksum verification, state machine. Code-level compliance wrapping probabilistic generation.
Structured JSON outputs with schema validation. Reduces parsing errors from 5% to less than 1%.
Each phase spawns fresh 200K token windows. No context pollution, no token accumulation.
Eight specialized phases, each with fresh context and deterministic validation
Researches requirements, analyzes technical constraints, queries pattern library
scout_report.json
Designs system architecture, file structure, test strategy from SCOUT_JSON
architecture.json
Implements code from ARCHITECTURE_JSON with 2-8 parallel workers
source code + build-tasks.json
Runs tests, self-heals failures, validates against requirements
test-report.md
Creates GitHub repo, commits code, pushes to remote
github.com/repo
Each phase spawns a fresh Claude instance via subprocess with rules injected at spawn time. Post-phase validators verify outputs exist, checksums detect unauthorized modifications, and BAML schemas enforce type safety. Generate probabilistically, validate deterministically.
Manage builds and monitor progress from your terminal
# Mission Control TUI
cf # Launch interactive terminal UI
# Daemon management
cfd start # Start the daemon
cfd status # Check daemon health
cfd logs <job-id> --follow # Stream build logs
cfd list # List active builds
cfd stop # Stop the daemonInstall takes less than 2 minutes
# Option A: npm (recommended)
npm install -g context-foundry
# Option B: pip
pip install context-foundry# In Claude Code, just ask in natural language:
"Use CF to build a mass text messaging app"
# Walk away. Come back to deployed code on GitHub.Generate probabilistically, validate deterministically