Using the MCP
This page is a simple guide to using Sneakers-PAM with an AI coding assistant. You don't need to know how any of it works under the hood. You just need to know what to type.
What is the MCP?
MCP stands for Model Context Protocol. It's a way for an AI assistant to talk to other tools. The Sneakers-PAM MCP lets your AI assistant ask Sneakers-PAM for the things it's allowed to see — like a password or a key — instead of you copying and pasting it by hand.
Think of it like this: the AI assistant doesn't have its own key to the vault. It has to ask you (or your team's rules) for permission every time, the same way a person would. Nothing happens in secret.
Before you start
You'll need Sneakers-PAM set up and your coding assistant connected to the MCP server. The exact setup steps will be added here once they're ready.
Things you can ask it to do
Once it's connected, you can just type plain sentences to your assistant. Here are some examples.
Find a secret
"Can you find the database password for the billing service?"
The assistant will look through Sneakers-PAM for a secret that matches and tell you what it found. It won't show you the actual password unless you ask it to, and unless you're allowed to see it.
See what's in a folder
"What secrets are in the 'staging' folder?"
This lists the names of the secrets, not their values. It's a safe way to look around.
Reveal a value
"Reveal the API key for the weather service so I can test it."
This is the one that needs the most care. Revealing a value is always written down in Sneakers-PAM's history, so there's a record of who asked and when. For a secret you can read, it just works, with no extra sign-in step once your assistant is signed in. If the secret is set to need an approval, an owner or approver says yes first (never you), or, when nobody else can, you confirm the task once in your browser. See who approves what.
Run something with a secret, without seeing it
"Run the deploy script using the deploy-bot credentials, but don't show me the password."
This is often the safest way to use a secret with an assistant. The password gets used behind the scenes, and you never actually see it typed out anywhere.
Rotate a credential it manages
"Rotate the local admin password for host-a and set it there."
For a secret the assistant is allowed to create and update (not just read), it can generate a fresh value, store it, and hand it back once so it can be set on the system. If setting it there fails partway, ask it to put the previous value back — that's a new version, never a rewrite of what came before.
Generate an SSH key
"Make a new Ed25519 SSH key called deploy-bot in the 'staging' folder."
Sneakers-PAM makes the key pair and stores it as a new SSH Key secret in one step. The assistant only gets the public key and its fingerprint, so the private key never appears in your chat. You can also ask for RSA 4096, RSA 2048 or ECDSA P-256. Put the public key on the systems the key signs in to.
Create a folder
"Create a folder called 'agents' under 'staging'."
The assistant can make a folder inside one it's allowed to write to. If you don't name a parent, it makes a top-level folder that belongs to you. An admin can restrict top-level folders to admins, and the assistant then tells you so. It can't share a folder: you do that in the staff app, and the assistant's token doesn't see a new top-level folder until then. See Top-level folders.
Attach a secret to another system
"This SSH key also needs to work on host-b — attach it there too."
One secret can be attached to more than one target. Attaching or detaching one never disturbs the others already attached.
A few things to keep in mind
- The assistant follows the same rules you do. It can't see or do anything you aren't allowed to see or do yourself.
- Asking for a secret to be revealed to an assistant is treated as extra sensitive. Expect it to need more care than just looking something up, and, for a secret set to need one, someone else's approval.
- When you can, ask it to run something with the secret instead of showing you the secret. That way the value never has to be typed out or saved anywhere, by you or the assistant.
- Everything is logged. If something looks wrong later, there's a record to check.
To understand what's happening behind the scenes, see Agents and the MCP and Approvals and requests.