Skip to content

usage.md

Latest commit

 

History

History
166 lines (123 loc) · 7.62 KB

usage.md

File metadata and controls

166 lines (123 loc) · 7.62 KB
Raw

Usage Examples

Table of Contents

Example 1: Generating a DOCX file

Word generation modes: ENABLE_WORD_ELEMENT_FILLING=false is the default mode. In this mode the LLM writes Python code and the backend executes it to generate the DOCX. This default mode supports images uploaded in the chat. ENABLE_WORD_ELEMENT_FILLING=true is an experimental structured mode where the model decides the logical order of document elements and their content while the backend builds the DOCX from a template instead of executing generated code. This experimental mode aims to replace code generation over time, but not all models perform well yet. The Results Summary 📊 benchmark below gives a practical idea of which tested models performed best in this mode.

This alpha version includes both DOCX generation approaches. The examples below show the kind of output GenFilesMCP can produce in practice.

Generating DOCX Example

This new version can include images embedded directly into the generated Word document, sourced from chat uploads. This allows for richer, visual content in the documents without relying on external links.

Generating DOCX Example

Your assistant can generate documents with one column or two columns for academic papers.

You can find results like this in the example\DOCX folder of the repository. Each document was exported manually as .pdf to be able to view the results in github, but you can find the original .docx files in the same folder.

Results Summary 📊

Scoring scale: 1 = full star, 0.5 = half star, 0 = empty star.

Criteria (brief):

  • C1 Image Use: correct and meaningful integration of provided images.
  • C2 Structure: logical flow and coherence.
  • C3 Word Elements: good use of title/sections/tables/lists/images/formatting.
  • C4 Depth: ability to develop the topic with complete, well-elaborated paragraphs (not short, shallow ones).
  • C5 First-Try Success: generated successfully on first attempt.
Models that support images
Model C1 C2 C3 C4 C5 Total Stars
GPT 4.1 mini010.50.513.0★★★☆☆
GPT 5.1 mini0.50.510.513.5★★★⯪☆
GPT 5.1 Codex mini0.510.5114.0★★★★☆
GPT 5.2110.5114.5★★★★⯪
Claude 3 Haiku010.50.513.0★★★☆☆
Claude Haiku 4.5111115.0★★★★★
Google Gemini 3 Flash Preview0.511103.5★★★⯪☆
Gemini 3 Pro Preview000000.0☆☆☆☆☆
Grok Code 4.1 Fast110.50.514.0★★★★☆
Kimi K2.5111115.0★★★★★
Mistral 14B 2512000000.0☆☆☆☆☆
Qwen3 VL 8B Thinking0.50.51013.0★★★☆☆
Models that do not support images
Model C1 C2 C3 C4 C5 Total Stars
Grok Code Fast 1N/A11114.0★★★★☆
DeepSeek V3.1 TerminusN/A11114.0★★★★☆
  • 🏆 Best overall: Claude Haiku 4.5 and Kimi K2.5 (5.0/5).
  • ✅ Acceptable performance (4.0 to <5.0): GPT 5.2 (4.5), GPT 5.1 Codex mini (4.0), Grok Code 4.1 Fast (4.0), Grok Code Fast 1 (4.0), DeepSeek V3.1 Terminus (4.0).
  • ⚠️ Weakest results: Gemini 3 Pro Preview (0.0) and Mistral 14B 2512 (0.0).
  • ℹ️ In this evaluation, C1 = 0 means the model failed to call the tool needed to identify attached images.
  • 🔎 Content factuality, hallucinations, and technical correctness were not evaluated here, because this benchmark focuses on structure, tool use, and first-pass execution; users must validate factual quality independently.

If your model does not have vision capabilities do not attach images in the chat, as the agent will not be able to see them and include them in the document.

Example 2: Generating a XLSX file

Generating XLSX Example 1

Open the generated file in Excel:

Generating XLSX Example 2

Example files: You can find example XLSX files in the example folder.

Example 3: Generating a PPTX file

In this example, another MCP server was used for web research and GenFiles MCP Server was used to generate a PowerPoint presentation:

Generating PPTX Example 1

Open the generated file in PowerPoint:

Generating PPTX Example 2

Example files: You can find example PPTX files in the example folder.

Example 4: Reviewing a DOCX file with comments

The review feature allows the agent to analyze uploaded documents and add structured comments for improvements.

Review Example 1

Review Example 2

Workflow:

  1. User uploads History_of_Neural_Nets_Summary.docx to the chat
  2. User requests a review with comments for corrections, grammar suggestions, and idea enhancements
  3. Agent calls the chat_context custom tool to retrieve file ID and name
  4. Agent uses the list_docx_elements tool to analyze the document structure
  5. Agent calls the review_docx tool to add comments to specific elements

Result:

DOCX Comments

Example files: Find the reviewed document in the example folder: History_of_Neural_Nets_Summary_69d1751b-577b-4329-beca-ac16db7acdbd_reviewed.docx

Generated using the GenFiles MCP Server and GPT-5 mini

The review functionality preserves the original formatting while adding structured comments