Give AI agents a human-in-the-loop API.

When an agent needs information, a file, a correction, or a decision, it creates a hosted Fangst request. The person responds, and the agent continues with structured data.

From API call to structured result.

  1. Create the requestSoftware or an AI agent defines the fields and receives a hosted URL.
  2. Share the URLA person opens the Fangst page and provides the requested input or files.
  3. Read the resultFangst returns structured data through the API or a signed webhook.

Let the agent choose the right human interaction.

Expose Fangst as REST or MCP tools. The agent can create a request when it reaches missing information during a run.

Create

POST https://api.fang.st/v1/requests
Authorization: Bearer <FANGST_API_TOKEN>
Content-Type: application/json

{
  "tool": "request_human_input",
  "arguments": {
    "title": "Provide supplier details",
    "fields": [
      {
        "key": "bank_account",
        "type": "text"
      },
      {
        "key": "documentation",
        "type": "file"
      }
    ]
  }
}

Completed result

{
  "status": "completed",
  "response": {
    "company_name": "Northstar AS"
  }
}

Choose the human step by what the agent is missing.

Human in the loop is not one generic pause. The agent should create the smallest interaction that resolves the current uncertainty, persist its request id, and resume only from the recorded result.

Input or files

Ask for facts or documents that are unavailable to the agent, then return typed fields and authenticated attachments.

Review and correction

Pre-fill generated or extracted values so a person can confirm correct fields and change only what is wrong.

Approval

Bind approve or reject to the exact action, object, and revision so a later decision cannot authorize changed work.

Use REST, MCP, polling, or webhooks.

Software uses the HTTP resources directly. Authorized AI agents can use the same API or intention-oriented Fangst MCP tools.

POST/v1/requestsCreate a hosted request and receive its URL.
GET/v1/requests/:idRead request status and configuration.
GET/v1/requests/:id/submissionsRead completed responses.
MCPrequest_human_inputLet an authorized agent create the request.
MCPget_human_inputLet the agent read the completed response.

Fangst runs the hosted interaction.

Hosted page

Responsive pages, clear sender context, validation, private or shared links, expiry, and submission limits.

Structured result

Typed values, authenticated file downloads, correlation metadata, timestamps, and stable response ids.

Reliable delivery

Polling, signed webhooks, retries with backoff, delivery history, idempotency, and replay.

Pay for completed interactions. No subscription.

Start with $1 free. Add funds from $10. Links, polling, webhooks, and abandoned requests do not use the main billable unit.

See pricing

Give your software a way to ask people.

Create the hosted interaction through Fangst and receive the structured result back.

Contact