runapi-mcp — agentic threat model
The runapi-mcp agent introduces elevated risk by acting as a bridge between Claude Code and external media/LLM generation APIs, creating a direct vector for credential theft, unauthorized API consumption, and financial exposure through model pricing and task execution tools.
OWASP AIVSS score rationale
| Autonomy of Action | 0.60 | |
| Goal-Driven Planning | 0.50 | |
| Self-Modification | 0.10 | |
| Dynamic Tool Use | 0.80 | |
| Persistent Memory | 0.30 | |
| Contextual Awareness | 0.60 | |
| Dynamic Identity | 0.70 | |
| Multi-Agent Interactions | 0.40 | |
| Non-Determinism | 0.70 | |
| Opacity & Reflexivity | 0.50 |
Scored with the canonical OWASP AIVSS formula (AIVSS calculator reference); agentic risk factors estimated from the agent’s described capabilities.
MAESTRO 7-layer threat model
Per-layer threats for this agent. Layers tagged “not certain from listing” are general, caveated commentary where the public description didn’t pin that layer.
Not certain from the listing — relies on external foundation models and Claude Code's base model. Threats include adversarial prompt injection via model browsing descriptions or task inputs that could hijack the agent's execution flow.
Not certain from the listing — data operations are primarily transient, handling API payloads, model pricing metadata, and generated media tasks. The primary threat is data exfiltration of generated assets or API keys passing through the context window.
The agent framework (Claude Code + RunAPI MCP server) is highly vulnerable to tool misuse and insecure tool integration. Malicious inputs could force the MCP server to execute unauthorized API calls, generate expensive media tasks, or leak active session credentials.
Deployment involves a local MCP server integrated with Claude Code making external network calls. This creates a network/credential surface where local API keys, environment variables, or session tokens could be exposed to the host or intercepted.
Not certain from the listing — there is no mention of built-in logging, guardrails, or cost-limiting observability to prevent budget exhaustion from runaway media generation tasks.
Not certain from the listing — lacks explicit security controls, identity management, or authorization policies to restrict which local users or parent processes can invoke the authenticated RunAPI endpoints.
Operates within the Claude Code ecosystem as an MCP plugin. It faces risks of trust abuse where other local tools or agents running in the same Claude session could exploit runapi-mcp to make external calls without explicit user consent.
MAESTRO — the 7-layer agentic threat-modeling framework (Cloud Security Alliance / Ken Huang).