Skip to content
Gasket

03Credentials · What can it reach?

Here is exactly how access works today. Your systems’ credentials stay with you: in your tool endpoints, or in your own agent’s code. When a Gasket agent calls a tool, Gasket checks the policy and then sends a signed request to your endpoint, which uses its own credentials. Gasket never stores or sees them.

Today

your agentholds STRIPE_KEY
│ asks: may I stripe.refund.create 240?
gasketnever sees the key
│ answers: authorized · rule 1
your agent → stripeuses its own key

Planned

your agentholds no secrets
│ asks
gasket vaultscoped, per action
│ short-lived access only when authorized
stripe—

Not built. Shown so you know where this is going, not as something you can use.

What this means in practice

  1. 01Tools hold the keysA Gasket-run agent reaches Stripe or your database only through your tool endpoints. Gasket calls them only for authorized tool calls, and signs every request so your endpoint can verify it came from Gasket.
  2. 02Your own agentsIf you run your own agent code, call POST /v1/actions inside the function that holds the credential, and only use it when the answer is authorized.
  3. 03Least privilege still mattersGive the agent's own credentials the narrowest scope the system allows, such as restricted Stripe keys or fine-grained GitHub tokens. Gasket adds a decision on top, not instead.
  4. 04Gasket's own keysGasket API keys are hashed with SHA-256 and shown once. They identify your workspace and environment, and grant nothing on other systems.
  5. 05With MCPAn AI connected over MCP asks Gasket and follows the answer, but Gasket can't physically block a tool it doesn't sit in front of. For hard enforcement, gate the tool in code.

PlannedA credential vault: store a system’s credentials in Gasket and have them injected, short-lived and scoped, only into authorized actions. Your agent would never hold the secret.

PlannedPer-agent credential bindings shown on the agent’s page, with which operations each credential may be used for.

See the roadmap for what’s next, and security for what’s true now.

← 02PoliciesWhat may it do?04 →ActionsWhat is it attempting?