What problem does it solve? Connecting an AI agent to external MCP servers introduces security and governance risks: unapproved network access, leaked credentials, and untrusted server responses. This Skill provides a provider-neutral adapter contract that enforces explicit approval, secure credential handling, and auditable evidence before any MCP connection is used. ## Core Features & Use Cases - Provider-Neutral Adapter Contract: Works with any host-exposed MCP transport without assuming a specific agent, shell, package manager, or configuration path. - Approval and Safety Gates: Requires explicit approval for new servers, network access, uploads, or credential scopes, and treats all server responses as untrusted input. - Redacted Evidence Recording: Logs server identity, tool calls, outputs, and failures as redacted evidence whenever they affect a project decision. - Use Case: When a research workflow needs an external MCP tool, the agent reads the capability registry, requests approval for the new server, keeps credentials in the host's secure store, and records redacted call evidence—or falls back to a local path if no adapter is available. ## Quick Start Use the native-mcp skill to check the capability registry and connect to an approved MCP server for this project.
