infinite_orchestrateLABSCO SUMMARY
Parallel generation with the bookkeeping done — uniqueness, sophistication levels, and a context ceiling.
The interesting part is not bulk generation but that it tracks what has already been generated. Each parallel agent is handed its own innovation dimension, later waves are told to go past the earlier ones, and generation stops on a context threshold instead of degrading into repetition. INFINITE mode means 'until context runs out', which is why context monitoring is a first-class tool here rather than an afterthought.
What it is
An orchestration server that runs the infinite agentic loop pattern for any kind of output. You hand it a specification — domain, output format, naming pattern, quality standards, innovation dimensions — and it coordinates parallel agents through waves of increasing sophistication.
What you get
- infinite_orchestrate — the main call: a specification, an output directory, and a mode of SINGLE, BATCH, or INFINITE
- wave_plan — plan the next wave against the work that already exists, a sophistication level, a target count, and a context budget
- agent_coordinate — run agents in parallel, each on a distinct innovation dimension, with duplicate concepts prevented across streams
- context_monitor — watch context capacity and shut down gracefully at a threshold rather than stopping mid-generation
- spec_validate — check a partial specification and fill in defaults before a wave is spent on it
- Evolution patterns of LINEAR, EXPONENTIAL, ADAPTIVE, or CREATIVE_BURST across domains from UI and CODE to DOCUMENTATION, RESEARCH, CONTENT, ANALYSIS, and DESIGN
- Per-domain validation rules covering syntax, semantics, functionality, and quality, with quality scoring on each iteration
Requirements
No account and no key. There is no published package: clone the repository, run npm install and npm run build, then point your client at the built dist/server.js and launch it with node over stdio.
Setup effort
Build from source — git clone https://github.com/gptcompany/universal-infinite-loop-mcp-server.git
USE CASES
01
Producing ten takes on a component and comparing them
Validate the specification first, then run it in batch mode so the variations come back coordinated rather than near-identical.
02
Keeping parallel agents from writing the same thing twice
Plan the wave against what already exists, then coordinate the agents with one innovation dimension each.
03
Running a long generation without falling off a context cliff
Set a capacity threshold and let the run close itself down cleanly instead of truncating mid-output.
04
Turning a vague brief into something the loop can actually run
Pass the partial specification and get back the filled-in version, with the output requirements and validation rules made explicit.
5 tools
Listed in project docs. Your client may show a different list.
wave_planagent_coordinatecontext_monitorspec_validateBEFORE YOU INSTALL
What to know first
Tools you get
5 tools named in the project's documentation.
Where it runs
The documented client runs the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
# Clone and install
git clone https://github.com/gptcompany/universal-infinite-loop-mcp-server.git
cd universal-infinite-loop-mcp-server
npm install
# Build
npm run build
# Run in development
npm run devMCP Clients
Claude Code
CLAUDE CODE · JSON
{
"mcpServers": {
"infinite-loop": {
"type": "stdio",
"command": "node",
"args": [
"<YOUR_PATH_TO_UNIVERSAL_INFINITE_LOOP_MCP>/dist/server.js"
],
"env": {}
}
}
}
