VectorOWL: Autonomous Systems Engineering

Before the math

Framework, language, or skill?

• VectorOWL is a research architecture / framework (runtime in development): OWL plus vectors plus MCP integration—it is not a replacement for Python, C++, or similar general-purpose languages.
• Single source of truth: the versioned OWL graph is where governed system and data-model semantics, characterization, traces, and evidence converge. Tools feed it; they should not be the silent second book of record.
• OWL is the formal ontology language for structure, constraints, and provenance in the graph.
• The real integration surface is Model Context Protocol (vectorowl-mcp, stdio): register it in Claude Desktop, Cursor, or any MCP host so tools and resources talk to the runtime. A bundled SKILL.md is only optional prose for assistants; it does not replace the MCP server. (Do not confuse this with the INCOSE “Model Characterization Pattern” for computational models—that is metadata and requirements on models; see framework · characterization.)

Everything below stays the same—this section only orients readers who hit the symbols first.

Before you paste

1. vectorowl-mcp is on your PATH, or change the line that sets "command" to the absolute path of your binary.
2. After the script prints “Done,” restart Cursor or Claude Desktop (whichever config you patched).

Cursor — ~/.cursor/mcp.json (Linux, macOS, WSL)

Registers the server for all workspaces. For one repo only, edit the script: set config_path to your project’s .cursor/mcp.json path.

python3 << 'EOF'
import json, os

config_path = os.path.expanduser("~/.cursor/mcp.json")
os.makedirs(os.path.dirname(config_path), exist_ok=True)

try:
    with open(config_path) as f:
        config = json.load(f)
except (FileNotFoundError, json.JSONDecodeError):
    config = {}

config.setdefault("mcpServers", {})["vectorowl-runtime"] = {
    "type": "stdio",
    "command": "vectorowl-mcp",
    "args": [],
    "env": {
        "VECTOROWL_LOG_LEVEL": "info"
    }
}

with open(config_path, "w") as f:
    json.dump(config, f, indent=2)

print("Done! Config written to:", config_path)
EOF

Claude Desktop — macOS (bash / zsh)

python3 << 'EOF'
import json, os

config_path = os.path.expanduser("~/Library/Application Support/Claude/claude_desktop_config.json")
os.makedirs(os.path.dirname(config_path), exist_ok=True)

try:
    with open(config_path) as f:
        config = json.load(f)
except (FileNotFoundError, json.JSONDecodeError):
    config = {}

config.setdefault("mcpServers", {})["vectorowl-runtime"] = {
    "type": "stdio",
    "command": "vectorowl-mcp",
    "args": [],
    "env": {
        "VECTOROWL_LOG_LEVEL": "info"
    }
}

with open(config_path, "w") as f:
    json.dump(config, f, indent=2)

print("Done! Config written to:", config_path)
EOF

Claude Desktop — Linux, WSL2, or Git Bash on Windows

python3 << 'EOF'
import json, os

config_path = os.path.expanduser("~/.config/Claude/claude_desktop_config.json")
os.makedirs(os.path.dirname(config_path), exist_ok=True)

try:
    with open(config_path) as f:
        config = json.load(f)
except (FileNotFoundError, json.JSONDecodeError):
    config = {}

config.setdefault("mcpServers", {})["vectorowl-runtime"] = {
    "type": "stdio",
    "command": "vectorowl-mcp",
    "args": [],
    "env": {
        "VECTOROWL_LOG_LEVEL": "info"
    }
}

with open(config_path, "w") as f:
    json.dump(config, f, indent=2)

print("Done! Config written to:", config_path)
EOF

Claude Desktop — Windows (PowerShell — paste as one block)

Writes %APPDATA%\Claude\claude_desktop_config.json. Uses a here-string so you do not need bash. If python is not found, try py -3 in place of python.

@'
import json
import os
from pathlib import Path

appdata = os.environ.get("APPDATA")
if not appdata:
    raise SystemExit("APPDATA is not set.")

config_path = Path(appdata) / "Claude" / "claude_desktop_config.json"
config_path.parent.mkdir(parents=True, exist_ok=True)

try:
    config = json.loads(config_path.read_text(encoding="utf-8"))
except (FileNotFoundError, json.JSONDecodeError):
    config = {}

config.setdefault("mcpServers", {})["vectorowl-runtime"] = {
    "type": "stdio",
    "command": "vectorowl-mcp",
    "args": [],
    "env": {
        "VECTOROWL_LOG_LEVEL": "info"
    }
}

config_path.write_text(json.dumps(config, indent=2), encoding="utf-8")
print("Done! Config written to:", config_path)
'@ | python -

Git Bash or WSL on Windows can use the Linux script above. If python - does not read stdin on your install, save the Python lines (without the PowerShell wrappers) into patch_claude_mcp.py and run python patch_claude_mcp.py.

Prefer raw JSON?

Copy the block from MBSE overview · Register the MCP server into the right file (~/.cursor/mcp.json vs Claude’s path). If the file already lists other MCP servers, add only the vectorowl-runtime object inside mcpServers.

Industry alignment

Model Characterization Pattern × VectorOWL

The MBSE community’s Model Characterization Pattern (same “MCP” acronym as in the MBSE wiki—not Anthropic’s Model Context Protocol) is the official S*Pattern for universal characterization and labeling of computational models: stakeholder and technical requirement groups, VVUQ, lifecycle, interoperability, and trusted reuse across supply chains—exactly as summarized in the v1.8.1 specification’s stated purposes. VectorOWL aligns with that intent as governed-graph substrate: OWL individuals and properties hold structured characterization; vectors support similarity search across model families and evidence; Anchors encode non-negotiable limits in review; Model Context Protocol (the AI tool protocol) connects authoring and execution tools so evidence stays attached to URIs. We do not claim a complete, out-of-the-box encoding of every configurable feature group in the pattern PDF—implementation evolves in the VectorOWL codebase.

Primary reference: Model Characterization Pattern v1.8.1 (PDF). Pattern hub: OMG MBSE Patterns. Product narrative: Framework · computational models.