For AI agents

A receipt tool
your agent can actually call.

Plug Ripceipt into Claude Code, the Anthropic Agent SDK, MoltBot, or your own homegrown loop. One endpoint, one stable schema, confidence on every field — exactly the shape agents need to decide what to do next without a human in the loop.

Get startedSee pricing →

What you get

The parts that actually
save you time.

Tool-call-shaped

A single POST that maps cleanly to a function-calling schema. No multi-step orchestration, no chained calls — drop it into your agent's tool list and it works the first time.

Deterministic JSON

Same fields, same types, same order, every call. Versioned at /v1 and stable. Your agent's downstream parser doesn't need defensive code for shape drift.

Confidence per field

Every value comes back with a 1–99 score. Agents branch on it: auto-approve high, escalate low, ask the user only when it actually matters.

How it works

Three steps,
start to file.

  1. 01

    Register the tool

    Add the Ripceipt extract tool to your agent — Claude Code, Anthropic SDK, OpenAI function-calling, MoltBot pipelines, LangChain. The OpenAPI spec is the source of truth for tool schemas.

  2. 02

    Agent calls, we extract

    Your agent decides a receipt needs parsing, calls the tool with a file or URL, and gets structured JSON back inside the same turn. No webhook plumbing required for synchronous flows.

  3. 03

    Agent acts on the data

    Categorize, file, reimburse, log to a sheet, or kick off the next step — your agent reasons over clean fields, not OCR slop. Confidence scores keep it honest.

FAQ

Questions for
for ai agents.

Does this work with Claude Code as a tool?
Yes. The /v1/receipts/upload endpoint maps to a single tool definition. Drop the OpenAPI spec into Claude Code's tool list (or hand-write the schema from /openapi.json) and the model can call it directly.
What about the Anthropic Agent SDK?
The Agent SDK's tool-use shape is a clean fit. Synchronous mode returns the structured receipt in the same response, so the agent can reason over it the next turn. Async mode is available if you want webhook-driven agent loops.
We use MoltBot for our bot pipelines — does it integrate?
Yes. MoltBot can call the REST endpoint directly or use the typed JS/Python SDKs. The deterministic JSON schema means your downstream MoltBot steps don't need shape-checking guards.
How do agents handle low-confidence extractions?
Each field returns a 1–99 confidence score. Common pattern: auto-process anything above your threshold, escalate below it. The agent reads the score and decides — no separate review API needed.
Do you publish a Model Context Protocol (MCP) server?
Not yet. The REST API + OpenAPI spec covers the agent use cases we've seen so far. MCP is on the roadmap; reach out if it's blocking your build and we can prioritize.
Is there a token-cheap response mode?
Use the fields query param to project only what your agent needs. A categorization-only agent can request `merchant,total,currency,confidence` and skip the full items array — cuts the response payload (and the agent's context cost) by an order of magnitude.

Related

Employee reimbursements
Same API, human approvers
Freelancers
For solo operators

A receipt tool
your agent can actually call.

Five dollars a month gets you fifty receipts and the full product. Cancel anytime.

Get startedSee pricing →