# rafsilva85/credit-optimizer-v5 [Health: Active]

**Category:** 🗣️ Conversational AI  
**Repository:** https://github.com/rafsilva85/credit-optimizer-v5  
**GitHub Stars:** 52  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/rafsilva85-credit-optimizer-v5

## Description
Save 30-75% on Manus AI credits through intelligent model routing, smart testing, and context hygiene. Audited across 53 scenarios with zero quality loss.

## Tools
Capabilities this server exposes over MCP:

- **analyze_prompt** — Analyze an AI agent prompt and return optimization recommendations.

Returns strategy, model recommendation, estimated credit savings,
quality impact assessment, and efficiency directives.
- **get_strategy_for_task** — Get the optimal strategy for a specific task type.
- **get_golden_rules** — Get the 10 Golden Rules for credit optimization with ZERO quality loss.

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

```json
"mcpServers": {
  "credit-optimizer-v5": {
    "command": "uvx",
    "args": ["mcp-credit-optimizer"]
  }
}
```

## Documentation

## What rafsilva85/credit-optimizer-v5 MCP server does

rafsilva85/credit-optimizer-v5 is a free MCP package for optimizing how Manus AI tasks consume credits. It evaluates a prompt before execution and selects an execution approach intended to preserve the required result while avoiding unnecessary cost. The project is designed for users who work with Manus AI and want an agent-accessible way to review task routing decisions.

Its scope is narrower than the paid Manus Power Stack. The MCP distribution contains the Credit Optimizer component, not the separate Fast Navigation component. It therefore addresses model and task routing, chat-mode identification, batching, and context handling rather than replacing browser navigation with a web-fetching toolkit.

## How it works

The optimizer first classifies the task by intent. The README identifies 12 intent categories, including coding, research, creative work, data tasks, and translation. Based on that classification and the apparent complexity, it selects Standard or Max mode, using Max only when the task needs it. It can also recognize simple question-answering, brainstorming, or translation requests that may be handled in Manus Chat Mode rather than Agent Mode.

Related work can be detected for batching, which reduces repeated overhead. The server also applies context hygiene by compressing accumulated context so later steps do not keep growing in cost. A quality veto rule is part of the design: when an optimization is judged likely to reduce quality, it is skipped instead of forced.

The reported evaluation covers 53 scenarios across code generation, creative writing, data analysis, research, translation, debugging, documentation, and mixed-intent tasks. Those results are claims from the project materials, not a guarantee for every prompt or Manus configuration.

## Setup and configuration

Install the package from PyPI and start it as a Python module:

```bash
pip install mcp-credit-optimizer
python -m mcp_credit_optimizer
```

A stdio MCP entry can use the Python command and module argument:

```json
{
  "mcpServers": {
    "credit-optimizer": {
      "command": "python",
      "args": ["-m", "mcp_credit_optimizer"]
    }
  }
}
```

It lists Claude Desktop, Cursor, Windsurf, Copilot, and other MCP-compatible clients as supported. With the MCP form, the optimizer is invoked when needed; it does not automatically affect every Manus prompt.

## Tools and capabilities

The rafsilva85/credit-optimizer-v5 MCP server provides capabilities for:

- Classifying prompts across multiple task intents.
- Selecting Standard or Max execution based on task complexity.
- Detecting requests that may fit Manus Chat Mode.
- Identifying related tasks that can be batched.
- Compressing accumulated context to limit token-cost growth.
- Rejecting an optimization when the quality veto condition applies.

## Limitations and notes

The MCP package is not the complete paid Power Stack. Fast Navigation, including the project’s HTTP fetching, authenticated cookie bridge, asynchronous parallel fetching, and disk caching features, is described as a Manus Skill and is not available through this MCP server.

The optimizer also does not operate invisibly on every prompt. The project specifically contrasts the MCP version, which saves credits when called, with the Manus Skill, which activates automatically. Reported savings and speed figures are project-reported measurements; actual results depend on task type, routing decisions, and Manus usage.

_Full upstream README: https://allmcps.com/mcp/rafsilva85-credit-optimizer-v5/readme_