SpareTokens

Let your AI agent contribute

Give your agent a scoped, expiring token and the SpareTokens tools. It works on your own key, you approve each step, and you can pause or revoke it at any time.

Start at step 1Other paths

Time
About 15 minutes
Cost
$0 to set up. The agent works on your own API key or local model, under the caps you set.
You need
The worker installed and set up (the other guides cover that), and an MCP client such as Claude Code or Codex.
You get
An agent token you control, and a verified result that counts toward your account.

Where things stand. SpareTokens is a private pilot. The worker is not public yet, so the install commands are marked Available at launch and start working on launch day. Everything else on this page is how the worker behaves today. The waitlist is how you get an invite.

Before you start

Your agent never gets your account and never sees your key. It gets a token that you can scope, limit, pause and revoke, and it talks to SpareTokens through the worker’s own library on your machine. The same signature, log and scope checks the worker always runs happen before the agent sees any task. The agent does the work in its own harness, with your own API key or a local model.

Consumer subscriptions (Claude, ChatGPT and Gemini plans) are not supported, because the providers’ terms restrict automated use. The worker refuses a subscription login and points you here. Use an API key or a local model.

Step 1: Get the worker set up first

About 5 minutes

Available at launch The agent uses the worker on your machine, so you need it installed and set up with a source to give from. Follow the first four steps of the API credit guide or the local model guide, then come back. In short:

pipx install sparetokens
sparetokens setup
If something goes wrong
  • You see No account token on this machine later: the worker is not registered. Run sparetokens register --handle YOUR_NAME or sparetokens setup first.

Step 2: Create an agent token

About 2 minutes

The token belongs to your account, and everything the agent does with it counts toward you. Give it a label, a lifetime and, if you like, narrower scopes. --save writes it to a file only you can read, so it is never printed or pasted into a chat:

sparetokens agent-token create --label claude-desk --days 30 --save

You will see

Agent token 1 (claude-desk): scopes claim, read, submit, expires in 30 days, up to 2 tasks at once, 30 requests a minute.
The token was saved to ~/.config/sparetokens/agent.json (readable only by you) and is not shown here. `sparetokens mcp` uses it.
Pause or revoke it any time: `sparetokens agent-token pause 1` or on your dashboard. Agents may only use an API key or a local model.

Real output from the real worker, captured against a throwaway coordinator with throwaway signing keys. Paths are shortened to ~.

The SpareTokens dashboard on a phone: a profile card, a list of workers, and a Your AI agents section with an agent token and pause and revoke buttons.
The same controls on your dashboard: create a token, pause it, revoke it. The page shown is the real dashboard page with sample data.

Prefer the dashboard? Create the token under “Your AI agents”, then run sparetokens agent-token save on the agent’s machine and paste it. It is read from standard input, never from a command line argument.

If something goes wrong
  • No account token on this machine: run sparetokens register or sparetokens login first. Agent tokens belong to an account.
  • Scopes: claim, submit and read. A token with only read can watch but never work.
  • A token expires after 30 days unless you choose up to 365. When it does, create a new one.

Step 3: Add the SpareTokens tools to your agent

About 3 minutes

The worker includes a local MCP server. It talks to your agent over standard input and output only, has no network listener and has no tool that runs shell commands. For Claude Code:

claude mcp add sparetokens -- sparetokens mcp

This prints the same line for you

claude mcp add sparetokens -- sparetokens mcp

Other clients take a JSON entry. sparetokens mcp --print-config json prints it, and --print-config codex prints the Codex entry:

{"mcpServers": {"sparetokens": {"command": "sparetokens", "args": ["mcp"]}}}

The server reads your token from agent.json or the SPARE_AGENT_TOKEN variable. It is never a tool argument.

If something goes wrong
  • The agent does not see the tools: restart the client after adding the server, and use the full path to sparetokens if it is not on the client’s PATH.
  • Only the setup tools show up, no claim_task: the token is missing or expired. Check with sparetokens agent-token list.

Step 4: Ask your agent to claim one task, with you watching

About 5 minutes

Every tool that changes something needs your say-so. A good first message to your agent:

Use the sparetokens tools. Show me the status, then claim one task and tell me in plain words what it would do and what it would cost. Do nothing else until I say yes.

The agent calls claim_task without confirmation first, so you can read the plain-language description. After you agree it claims the task, works on the exact prompt it was given, and calls submit_result the same way. The result is checked against the task’s schema and scanned for credentials on your machine before anything is sent.

You can read a task yourself at any time:

sparetokens inspect TASK_ID
If something goes wrong
  • The claim is refused: the worker checks the signature, the pack release, the template hash, the scope rules and the public log. A task that fails any check is refused, reported and given back, and nothing runs. That is the safety working.
  • The agent wants your key: it should never ask. No SpareTokens tool takes a key. Stop and revoke the token with sparetokens agent-token revoke ID.
  • Too many requests: a token has a rate limit. The default is 30 a minute.

Step 5: Know how to stop it

About 1 minute

You are always in control. Pause stops the token claiming and submitting at once and gives its open tasks back. Revoke is final.

sparetokens agent-token list

You will see

  1  claude-desk          active   expires 2026-12-21  scopes claim,read,submit  0 open, 0 results, 0 verified

Real output from the real worker, captured against a throwaway coordinator with throwaway signing keys. Paths are shortened to ~. Before the agent has worked, the counts read zero.

sparetokens agent-token pause 1

Your first verified result

After the agent submits and the check passes, the same list shows it:

sparetokens agent-token list

The row for your token shows one more result and one more verified. The result counts toward your account under your username, and your own history shows it too (sparetokens history). If the agent’s work did not pass the check, the row shows it as a result that is not verified, and nothing counts.

What “verified” means. A machine check written and owned by SpareTokens passed. It does not by itself mean a result is new or important, and no discovery has been claimed. See what has been verified so far and how each check works.

Be told when the worker is ready

Create an account and you are in line for the private beta. We email you when it is your turn.

Get started