Discover configured MCP servers, skills and hooks with Oktsec Node. Give your team a clear starting point for reviewing what agents use and the access those components bring.
A server connects to systems. A skill carries instructions. A hook can run on its own. Each needs a different review.
The client configuration declares a server command, arguments and credential-variable names.
Review the server’s permissions before deciding which tools the agent may call.
The skill declares review instructions and may bundle executable files or specify tools.
Inspect the instructions and bundled code with Aguara before approving the skill.
A configured event starts a command, without waiting for the model to choose a tool.
Review the hook’s command and inherited access. Confirm that unattended execution is intended.
Illustrative entries, not a customer snapshot. Available fields depend on the client, configuration and Node version. Configuration does not prove execution.
A useful inventory shows where the inspection stopped, as well as what it found.
Review its source and declared capabilities. Confirm who uses it and which permissions it receives.
Resolve access or scan limits before treating the inventory as complete.
Verify the version separately. A name or file timestamp is not evidence of the installed release.
Discovery reads supported local configurations. It is not a complete inventory of browser AI or SaaS agents.
Check client coverageUse Aguara-powered inspection to review supported source and instructions.
Limit supported routed actions through tools, parameters and runtime policy.
Review system and policy state. Confirm inventory exchange for your Node and Cloud versions.
AI agent discovery identifies the clients, tools and capabilities present in an environment. Oktsec Node reads supported local configurations for MCP servers and agent capabilities. The result supports component review and policy decisions; configuration does not prove execution.
MCP discovery records clients, configuration locations, server commands, arguments and environment-variable names. Supported capability records include kind, source, execution class, declared tools and manifest hash, with origin and version when available. Coverage varies by client and Node version.
No. Local configuration discovery does not establish a complete inventory of browser-based AI, SaaS agents or unsupported clients. Confirm the machines, users, configuration sources and versions included in the evaluation.
Unreadable configurations are reported separately from empty findings. Partial capability scans identify incomplete inspection. Missing versions and origins remain unknown. A file modification timestamp is not a last-execution timestamp.
The MCP inventory projection retains environment-variable names and drops their values. Commands and arguments can still contain sensitive details and need review before sharing. The capability projection omits instruction prose, absolute paths and hook commands, retaining a manifest hash.
Discovery provides observations. Enforcing an action requires a configured Control integration. Signal’s external monitoring tied to Node inventory is in development. Confirm the inventory exchange supported by your Node and Cloud versions.
Bring a representative environment. We’ll compare discovered components with the tools your team actually uses.
