For agents

An agent can hand over a credential without being able to read it back

X-TKN is callable by an agent today: a sandbox key with no signup, a public OpenAPI spec, and refusals that name the plan which would grant the capability. The MCP server is in build, and it will ship without a read tool.

Read the API docs See pricing
No signup
Sandbox key, 100 operations, 24 hours
402
Refusals an agent can act on
No read tool
Now, and after MCP ships
One-time handoff
Retrieval link
x-tkn.com/s/#7f3a9c1e…
one read expires in 24h unread
Encrypted before it left the sender
We hold ciphertext; the key stays in the link
Waiting for the one read…
Available now

The API is the agent interface

None of this is waiting on the MCP server.

  • An unauthenticated sandbox endpoint returns a working API key. No signup, no inbox, no CAPTCHA.
  • A public OpenAPI spec at https://x-tkn.com/openapi.json.
  • A paid capability refuses with 402, the capability name, the cheapest plan that grants it, and a checkout link a human can click.
  • Quota position on every metered response through x-rate-limit, x-rate-usage and x-rate-ceiling, including on the 429.
  • Passing your quota does not kill the integration: you are served to twice it, with three emails on the way.
In build

Three tools, and the one we are not shipping

The MCP server is specified and not yet built. When it ships it will expose create_secret, check_status and revoke_secret.

  • There is no read_secret tool, and there will not be one.
  • The agent creates the handoff. The human, or the target system, consumes it.
  • An agent cannot exfiltrate through a tool it does not have.

We are second to market, so here is the precise difference

Vaulted ships an MCP server today. It is free, needs no account, installs with npx, and its source defines create_secret, view_secret, check_status and list_secrets. We do not have a shipped MCP server, and claiming otherwise would be exactly the kind of thing this site exists not to do.

The difference is view_secret. Their source implements it to decrypt a secret and hand it back to the agent, and the npm release, 0.1.6, still answers "not implemented yet" for it. Ours has no equivalent and is not designed to acquire one. If your threat model includes the agent itself, whether that is prompt injection, a compromised tool chain, or a transcript that outlives the task, then that is the distinction to weigh.

Two of our three tool names are also theirs, so the lists will be read side by side. That is the comparison we would rather you made than one we wrote for you.

Try it first

Call it before you commit to it

A sandbox key costs nothing and needs no account.

Read the API docs Compare with Vaulted