nhartman000/dsvm-auto13-mcp
If you are the rightful owner of dsvm-auto13-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.
A geometry-aware reasoning and machine-to-machine communication module for dynamic systems analysis using the Model Context Protocol (MCP).
Niché Machine Language + DSVM/Auto13/Exploder MCP Server
A geometry-aware reasoning and machine-to-machine communication module for dynamic systems analysis using the Model Context Protocol (MCP).
Overview
This project implements a toolstack for reasoning about dynamic systems using the Niché machine language and three core components:
- Niché Machine Language - Emoji-based machine language for LLM-to-LLM and machine-to-machine communication
- DSVM (Dynamic Systems Variable Mapping) - Maps emoji-based TOTE groups into geometric models
- Auto13 - Meta-screen/evaluator for DSVM outputs
- Exploder - TOTE expansion engine for higher fidelity
Architecture
The system works as follows:
- LLM Client converts natural language queries into emoji-based TOTE groups
- DSVM Engine maps these groups into geometry and computes 4D context (relevance scores)
- Auto13 Engine evaluates the model's usefulness and suggests improvements
- Exploder Engine can expand coarse actions into finer sub-actions when needed
- MCP Server exposes these as tools via the MCP protocol
Installation
cd python
pip install -r requirements.txt
Usage
Running the MCP Server
cd python
PYTHONPATH=/home/deepsleep_r/projects/dsvm-auto13-mcp/python python -m dsvm_auto13_mcp.server
Running Tests
cd python
PYTHONPATH=/home/deepsleep_r/projects/dsvm-auto13-mcp/python python -m pytest dsvm_auto13_mcp/tests/ -v
Testing the Engines
PYTHONPATH=/home/deepsleep_r/projects/dsvm-auto13-mcp/python python test_engines.py
MCP Tools
The server exposes three tools:
dsvm_auto_map_with_context- Map emoji TOTE groups into geometry with 4D relevance contextauto13_screen_model- Screen an existing DSVM model dict and score its usefulnessexploder_expand_totes- Expand a coarse TOTE emoji group into subtotes for higher fidelity
Example Usage
from dsvm_auto13_mcp.dsvm import DSVMEngine
from dsvm_auto13_mcp.auto13 import Auto13Engine
# Create engines
dsvm = DSVMEngine()
auto13 = Auto13Engine()
# Map TOTE groups to geometry
model = dsvm.map_with_context(
dataset_ref="walk_to_store_case1",
query="Analyze walking to store and back",
raw_tote_groups=[
["🚶♂️", "🏠", "🏬"], # walk to store
["🧾", "💳"], # pay at store
["🚶♂️", "🏬", "🏠"] # walk back home
]
)
# Screen the model
screening = auto13.screen(model, screening_mode="normal")
print(f"Decision: {screening.decision}, Score: {screening.usefulness_score}")
Niché LLM-to-LLM Demo
The Niché machine language includes an LLM-to-LLM demo that shows how one model can:
- Compile natural language into a compact Niché machine language string
- Send only that compact string over a channel (no raw text)
- Have another model parse the compact Niché message back into a structured expression
Run the demo:
cd python
python demo_llm_to_llm_niche.py
You should see output similar to:
LLM A: natural language input:
Turn on the left power circuit, monitor current, and trigger an alert if the load exceeds safe limits.
Channel payload (compact Niché machine language):
28BF3BE112CE8602:NICHE_E1E6AC29:GEOM_69E9C672C61F:❓⚡🌊∧🔀⚡🌊∧🔀
Simulating LLM B receiving and interpreting the compact message...
LLM B: parsed Niché message:
machine_symbol: NICHE_E1E6AC29
geometry_signature: GEOM_69E9C672C61F
communication_hash: 28BF3BE112CE8602
semantic_stability: 1.000
emoji_sequence: ❓⚡🌊∧🔀⚡🌊∧🔀
You can pass your own text:
cd python
python demo_llm_to_llm_niche.py "Route power through backup circuit and raise an alert if voltage drops below threshold."
Internally, the demo uses the Niché compiler v2 helpers to go from natural language → compact Niché message (compile_for_llms in LLM_DIRECT mode) and back (parse_llm_message from the compact format).
Project Structure
python/
├── dsvm_auto13_mcp/
│ ├── __init__.py # Package exports
│ ├── server.py # MCP server implementation
│ ├── dsvm.py # DSVM engine
│ ├── auto13.py # Auto13 screening engine
│ ├── exploder.py # Exploder expansion engine
│ └── tests/ # Test suite
├── requirements.txt # Dependencies
└── setup.py # Package configuration
Development
This project follows the specification defined in the README documentation and implements all core concepts:
- TOTE groups as emoji clusters
- Time as the first dimension
- Dimension based on directional complexity
- Geometric mapping with polyhedra
- 4D context as relevance scores
- Auto13 screening with retry constraints
- Exploder expansion for fidelity improvement
License
MIT License - See LICENSE file for details.
GitHub Command Line (gh)
If you’re looking for the GitHub command line tool, it’s called “GitHub CLI” and its executable is gh.
Where to get it:
- Official site: https://cli.github.com/
- Documentation: https://cli.github.com/manual/
Install instructions (pick your OS):
- macOS (Homebrew):
- brew install gh
- Windows (winget or MSI):
- winget install --id GitHub.cli -e
- Or download the MSI from https://github.com/cli/cli/releases
- Debian/Ubuntu:
- type -p curl >/dev/null || sudo apt update && sudo apt install -y curl
- curl -fsSL https://cli.github.com/packages/githubcli-archive-keyring.gpg | sudo dd of=/usr/share/keyrings/githubcli-archive-keyring.gpg
- sudo chmod go+r /usr/share/keyrings/githubcli-archive-keyring.gpg
- echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main" | sudo tee /etc/apt/sources.list.d/github-cli.list > /dev/null
- sudo apt update && sudo apt install gh
- Fedora:
- sudo dnf install 'dnf-command(copr)'
- sudo dnf copr enable @github-cli/gh -y
- sudo dnf install gh -y
- Arch Linux:
- pacman -S gh
Quick start:
- Verify installation: gh --version
- Authenticate with GitHub: gh auth login
- Choose GitHub.com, HTTPS, and “Login with a web browser” (recommended)
Common commands:
- Clone this repo: gh repo clone dsvm-ai/dsvm-auto13-mcp
- Create an issue: gh issue create --title "Title" --body "Description"
- List issues: gh issue list
- Create a PR: gh pr create --fill
- Check PR status: gh pr status
Tip: You can also run these within the python subfolder by setting the working directory accordingly, but gh operates on the repository root by default.