# PromptSpend [Health: Active]

**Category:** 💰 Finance & Fintech  
**Repository:** https://github.com/AndrewAvery7/promptspend  
**GitHub Stars:** 1  
**npm Downloads (last month):** 1332  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/promptspend

## Description
LLM pricing where every rate carries its source and the date it was last confirmed.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "promptspend": {
    "command": "npx",
    "args": ["-y","@promptspend/mcp"]
  }
}
```

## Documentation

## What PromptSpend does

PromptSpend gives AI agents access to an LLM pricing catalog designed for cost analysis rather than a static list of rates. Each pricing entry carries provenance and the date it was last checked, allowing estimates to be reviewed against the vendor source. The catalog covers model prices, provider free-tier information, and the assumptions used in calculations.

The main use is planning an AI feature or workload. A developer can estimate spending for an interaction, compare several models, inspect per-user and margin figures, and create a cost-audit receipt that summarizes the visible workload. Results identify major cost drivers and can suggest lower-cost models to test without treating price as evidence of equivalent quality.

## How it works

A scheduled GitHub Action refreshes the catalog every morning from independent sources. PromptSpend combines those sources using an explicit trust order, applies sanity checks, and either commits the refreshed result or opens a pull request for human review. Capture rules operate at the model-family level, so new versions can be included without changing application code for each release.

Calculations separate input and output pricing. They can account for repeated chat history, model-specific tokenization, cache writes, long-context pricing tiers, hidden reasoning tokens, promotional-rate dates, and output ceilings. Assumptions and exclusions are shown with the result instead of being hidden in the calculation.

Free-tier records include provider rules, limits, card requirements, and whether free usage trains models. These facts are quoted from vendor pages with links and dates, then checked during the same morning refresh process.

## Tools and capabilities

- Estimate monthly, yearly, per-user, and margin-related costs for a described interaction.
- Compare up to four models side by side for a workload.
- Browse a sortable catalog with provider, source, and verification details.
- Produce a cost-audit receipt containing workload estimates, leading cost drivers, and candidate cheaper models.
- Review free-tier conditions across providers.
- Inspect pricing provenance, pipeline status, and flagged data.

## Limitations and notes

Estimates are not a substitute for measuring production traffic. Exact token counting is available for OpenAI-family models using their tokenizer; other model families use a clearly labeled calibrated ratio. Price does not establish response quality, so suggested alternatives require testing.

Caching is disabled by default because cache writes are billed separately. Promotional prices are applied only within their stated date window. Requests that exceed a model's context window or output limit are flagged as infeasible rather than assigned a misleading price.

Price alerts are opt-in and depend on a configured deployment. Available delivery methods include an Atom feed, repository pull requests, browser push, and an email digest. Email subscriptions store the address, followed models, and signup date; browser push uses the browser-issued subscription URL.

_Full upstream README: https://allmcps.com/mcp/promptspend/readme_