For developers

Start with an explicit action contract.

Use local policy evaluation to test a proposed workflow. Enforce any permission at the system that actually owns the credentials and performs the action.

1. Connect one local tool

In an MCP client that accepts the configuration below, add this server. Other clients may use a different settings format; use the same command and arguments. Node.js and npm must be installed.

{
  "mcpServers": {
    "dingdawg-agent-spend-policy": {
      "command": "npx",
      "args": [
        "-y",
        "@dingdawg/agent-spend-policy-mcp"
      ]
    }
  }
}

No DingDawg account or payment credentials are needed for this local evaluator. The first run downloads the package through npm.

2. Check a proposed action

Choose evaluate_spend_policy in your client. Supply the evaluation time, policy, proposed action and already-spent amount. Use integer micro-unit strings for money, as the package schema requires.

Read the input contract

3. Understand the result

ELIGIBLE: the supplied inputs meet the policy. DENY: a rule blocks them. STEP_UP: additional approval is required.

Read the reason code to understand the result. None of these outcomes submits a payment.

Keep the first setup small

Operate this evaluator through your MCP client; there is no separate hosted service to maintain. Your client starts and stops the local process. Removing its server configuration disconnects it.

The caller supplies the time and spending state. Use it for dry runs first. For a repeatable deployment, pin the package version you have tested; review changes before upgrading.

Evaluate loop intentions

Loop stores recurring intentions and evaluates them manually. Its schedule is metadata. It does not dispatch tools or run a payment connector.

Loop contract and installation

Explore the full catalog

Existing packages cover different jobs and have different limitations. Check the published version and its README instead of assuming one universal guarantee.

Browse packages

Before connecting money movement

Keep identity, policy evaluation, execution authorization and provider outcome separate. Use trusted state, exact-payload approvals, atomic limits, one-use identifiers and durable evidence at the execution boundary.

Pending or unknown provider outcomes require reconciliation. A deterministic release gate cannot turn an unknown transaction into a confirmed one.

Discuss an integration