Context Foundry
Spawn recursive Claude workers from an MCP server to autonomously implement features and rerun tests until a self-healing loop passes.
Overview
Context Foundry is a Build-phase MCP server that recursively spawns Claude sub-agents for autonomous project builds with self-healing test loops.
What is this MCP server?
- Recursive Claude spawning for autonomous multi-step project builds
- Self-healing test loops that re-delegate until tests pass
- PyPI package context-foundry version 2.2.1 with stdio MCP transport
- Requires ANTHROPIC_API_KEY for delegation features against the Claude API
- GitHub-hosted context-foundry project aimed at long-horizon agent builds
- Stdio transport with required secret ANTHROPIC_API_KEY for delegation
Community signal: 19 GitHub stars.
What problem does it solve?
One-shot agent sessions stall on multi-file features because there is no structured way to delegate subtasks and recover from failing tests automatically.
Who is it for?
Experienced solo builders with Anthropic API budget who want orchestrated multi-agent coding on a defined codebase.
Skip if: Beginners uncomfortable with API costs, unreviewed autonomous edits, or teams without automated tests to anchor the healing loop.
What do I get? / Deliverables
After setup, Context Foundry can loop delegated Claude workers against your repo until tests pass, reducing manual re-prompt churn during builds.
- Delegated multi-step implementation runs across recursive Claude workers
- Iterated test-fix cycles until the configured test suite succeeds
Recommended MCP Servers
Journey fit
Canonical shelf is Build → agent-tooling because the server orchestrates delegated Claude runs to construct software, not to operate production infra. Agent-tooling captures recursive spawning, delegation, and autonomous build loops rather than a single-shot codegen snippet.
How it compares
Recursive agent orchestration MCP, not a single static coding skill or passive documentation lookup.
Common Questions / FAQ
Who is Context Foundry for?
Solo and indie developers who already use Claude API access and want MCP-driven recursive agents to build and test scoped projects.
When should I use Context Foundry?
Use it in build/agent-tooling when you have tests in place and want autonomous delegation plus retry loops instead of manual multi-chat coordination.
How do I add Context Foundry to my agent?
Install context-foundry from PyPI, set ANTHROPIC_API_KEY, register the stdio MCP server in your client, and run it against a repo with a clear build and test command.