Preparing your workspace
Connecting MemonaIQ servicesPreparing your workspace
Connecting MemonaIQ servicesAudit MCP server configurations, tools, schemas, authorization policies, and dangerous capability combinations against the official MCP 2026-07-28 specification.
As LLMs integrate with local workstations and enterprise services via the Model Context Protocol (MCP), server configurations and tool definitions define the security perimeter. The MCP Security Auditor evaluates configuration files, stdio commands, environment variables, tool JSON schemas, and authorization boundaries entirely in your browser without executing tools or transmitting data.
Individual MCP tools may appear benign in isolation, but when combined within the same client session, they can create severe emergent risks. For example, pairing a local file reading tool with an outbound HTTP request tool creates a potential data exfiltration path under prompt injection. The auditor analyzes your multi-server capability surface and highlights hazardous pairings.
The MCP 2026-07-28 specification establishes strict requirements for HTTP transport authorization: clients must provide RFC 8707 Resource Indicators during token acquisition, and token passthrough is strictly prohibited. The auditor flags architectures that forward ambient tokens to downstream servers and recommends RFC 8693 Token Exchange patterns.
Declared tool annotations such as readOnlyHint: true or destructiveHint: false are metadata provided by the server. In accordance with official MCP guidance, the auditor surfaces these annotations as declared behaviors rather than verified security guarantees.
The MCP Security Auditor is a defensive, local-first static analysis tool that audits MCP server configurations, tool definitions, schemas, resources, prompts, transport settings, and authorization policies against official MCP specification requirements.
No. The auditor strictly inspects JSON/YAML configurations, schemas, and list payloads statically in your browser. It never executes tools, runs shell commands, launches background processes, or sends requests to third-party endpoints.
The auditor references the official Model Context Protocol revision (2026-07-28), which introduced a stateless protocol core, Mcp-Method routing headers, and hardened authorization boundaries, while maintaining compatibility analysis for legacy revisions (2025-11-25, 2025-06-18).
Forwarding ambient user authorization tokens directly to third-party MCP servers violates resource-server boundaries and enables audience confusion attacks. The 2026-07-28 MCP specification explicitly prohibits token passthrough, requiring downscoped OAuth Token Exchange.
No. Official MCP security guidance specifies that tool annotations represent declared server behavior and should be treated as untrusted unless the server itself is fully verified.
High-impact combinations occur when an MCP setup exposes multiple distinct capabilities simultaneously—such as local filesystem read access combined with outbound network access, which introduces potential data exfiltration risk under prompt injection.
Discover other utilities in the MemonaIQ ecosystem designed to help you analyze and optimize your digital infrastructure.
Inspect generated or copied code for leaked secrets, unsafe patterns, configuration mistakes, reliability issues, and deployment hazards directly in your browser.
Count o200k_base and cl100k_base tokens locally, inspect context usage, compare tokenizer efficiency, estimate API cost, and truncate prompts deterministically.
Compare OpenAPI 3.0 and 3.1 specifications semantically. Detect breaking changes across endpoints, parameters, request bodies, responses, enums, nullability, and security requirements.
