AI Agent MCP Tool for Local Document, PDF, PNG/Image, QR, RAG, CSV & File Processing

English · ภาษาไทย · Try the original browser tools · Tool reference · Roadmap

Since publishing our free browser-based PDF tools, image converter and compressor, Thai-English OCR, DOCX converter, QR generator, AI token/RAG utilities, and CSV/JSON/Excel tools, many people have asked how they could let their AI or coding agents access that tool bundle and do the work for them. DJAI Tools MCP is our answer: a free, open-source, local-first Model Context Protocol server that gives AI agents carefully controlled file-processing capabilities on the user's own computer.

The goal is to help ChatGPT-compatible agents, Claude, Codex, Cursor, VS Code agents, and other MCP clients prepare context, count tokens, create RAG chunks, clean CSV/JSON data, generate QR files, and—through staged releases—process PDF, DOCX, OCR, PNG, JPG, WebP, images, and spreadsheets without sending private documents to DJAI.

Current release scope: v0.1.0 contains eight tested text, RAG, CSV, JSON, and QR tools. PDF conversion, DOCX, OCR, PNG/JPG image processing, and XLSX tools are published in our existing browser-tool collection and are planned for later MCP phases; they are not falsely advertised as working MCP commands in this release.

Why we built this

Browser tools are useful when a person wants to open a page and process a file. An MCP server serves a different workflow: the user asks an agent to prepare context, clean data, generate an artifact, or perform a multi-step task, and the agent calls narrowly described tools on the user's behalf.

This project is designed to provide:

• Local privacy: files are read and processed on the user's machine.
• Agent access: MCP-compatible agents can discover and call enabled tools.
• Controlled permissions: the server reads and writes only inside allowed directories.
• Repeatable workflows: structured results report paths, sizes, counts, warnings, and processing time.
• Smaller tool menus: profiles register only the capabilities an agent needs.
• Open development: source code is free under the MIT License and welcomes community review.
• No DJAI processing bill: users provide their own compute and storage; DJAI does not host document processing.

Working MCP tools in v0.1.0

Capability	MCP tool	What the agent can achieve
AI token counter	djai_ai_count_tokens	Measure prompt, document, code, and context size using common tokenizer encodings
AI context optimizer	djai_ai_optimize_context	Remove repeated page furniture, broken prose wrapping, and excess whitespace
RAG chunk generator	djai_ai_chunk_for_rag	Create deterministic token-aware JSONL chunks with configurable overlap
Prompt/context packager	djai_ai_package_prompt	Combine local source files with Markdown or XML boundaries for an AI agent
CSV to JSON	djai_data_csv_to_json	Convert header-based CSV rows into a formatted JSON array
JSON to CSV	djai_data_json_to_csv	Convert an array of flat JSON objects into spreadsheet-friendly CSV
CSV cleaner	djai_data_clean_csv	Trim fields, remove empty rows, and remove exact duplicates
QR SVG generator	djai_qr_generate	Generate a local SVG QR code from text, URLs, Wi-Fi payloads, or identifiers

Planned phases cover merge/split/rotate/watermark PDF, PDF text and AI Markdown extraction, local Thai-English OCR, DOCX conversion, PNG/JPG/WebP image conversion and compression, PDF-to-image and image-to-PDF processing, and CSV/XLSX workflows. See ROADMAP.md.

Example agent workflows

• “Count the tokens in these requirements and package only the relevant sources for Codex.”
• “Clean this CSV, remove exact duplicates, convert it to JSON, and report the row counts.”
• “Prepare this extracted document text as 700-token RAG chunks with 100-token overlap.”
• “Generate a high-error-correction SVG QR code for this URL.”
• Later releases: “OCR these Thai scanned pages, clean the text, and prepare a RAG knowledge base.”
• Later releases: “Merge these PDFs, add page numbers, compress the output, and preserve the originals.”

Privacy and security model

The server uses local MCP over stdio; it is not a hosted upload API.

• Read root defaults to the current working directory.
• Write root defaults to ./djai-output.
• Existing files are not overwritten unless --overwrite is explicitly enabled.
• Real paths and symlinks are checked before file access.
• Defaults limit files to 25 MB, packaged batches to 20 files/50 MB, inline text to 100,000 characters, and CSV/JSON to 100,000 rows.
• Remote URL fetching and shell commands built from model input are not implemented.
• Every successful operation reports the files it created and modifiedInput: false.

An MCP process still inherits the permissions of the account that starts it. Authorize narrow folders and read the complete security model.

Download, install, and build

Requires Node.js 22.13 or newer and npm.

git clone https://github.com/lovecatisgood-sudo/AI_Agent_MCP_Tool_for_Multi_Docu_PDF_PNG_QR_File_Conversion_Processing_Capability.git
cd AI_Agent_MCP_Tool_for_Multi_Docu_PDF_PNG_QR_File_Conversion_Processing_Capability
npm ci
npm run build
node dist/index.js --version

Run the server from source:

node dist/index.js --profile core \
  --allow-read /absolute/path/to/documents \
  --allow-write /absolute/path/to/djai-output

After the npm package is published, the intended one-command installation is:

npx -y @djai/tools-mcp --profile core \
  --allow-read /absolute/path/to/documents \
  --allow-write /absolute/path/to/djai-output

Connect an MCP-capable agent

For a source checkout, use the absolute path to the built server:

{
  "mcpServers": {
    "djai-tools": {
      "command": "node",
      "args": [
        "/absolute/path/to/AI_Agent_MCP_Tool_for_Multi_Docu_PDF_PNG_QR_File_Conversion_Processing_Capability/dist/index.js",
        "--profile",
        "core",
        "--allow-read",
        "/absolute/path/to/documents",
        "--allow-write",
        "/absolute/path/to/djai-output"
      ]
    }
  }
}

Use absolute paths because desktop agents may start the process from an unexpected directory. Client-specific examples for Claude Desktop, Codex, Cursor, and VS Code are in examples/.

Profiles and exact tool selection

node dist/index.js --profile core
node dist/index.js --profile ai-documents
node dist/index.js --profile data
node dist/index.js --profile qr
node dist/index.js --profile full

# Exact selectors override the profile
node dist/index.js --tools ai.count_tokens,ai.chunk_for_rag

core registers seven broadly useful tools; smaller profiles reduce agent context and selection ambiguity. Read profiles and the complete tool reference.

Original free browser tools

The MCP project grows from the earlier Free PDF, Image, OCR, DOCX, QR, AI and Data Tools. Those applications already provide live, private browser workflows for PDF, image resizing/conversion/compression, Thai-English OCR, documents, QR, AI context/RAG, CSV, JSON, and Excel.

Browser collection	English	ไทย
All tools	Open	เปิดเครื่องมือ
PDF tools	Open	เปิด
Image tools	Open	เปิด
Document and OCR	Open	เปิด
AI token and RAG	Open	เปิด
CSV/JSON/Excel	Open	เปิด
QR generator	Open	เปิด

About DJAI Academy, Siamese Cat Dev, and Siamese Cat Cafe

• DJAI Academy develops practical technology education, software services, AI-agent integrations, and free digital tools for people and businesses.
• Siamese Cat Dev is the independent developer and product builder behind this open-source MCP release and the original browser-tool collection.
• Siamese Cat Cafe is the connected community business helping support cat adoption and independent creative development. Visitors can also find the cafe on Google.

Stars, contributions, honest feedback, visits to the cafe, and genuine reviews from people who have visited all help sustain the software and community mission.

Follow Siamese Cat Dev

• GitHub: @lovecatisgood-sudo
• Developer profile: Siamese Cat Dev
• X: @siamesecatdev
• Instagram: @djcatdev
• Facebook: Siamese Cat Dev
• Website: DJAI Academy

Development and contributions

npm run check
npm test
npm run build
npm pack --dry-run

Use only synthetic fixtures—never commit personal, proprietary, confidential, or credential-bearing documents. Read CONTRIBUTING.md, SUPPORT.md, SECURITY.md, and the Code of Conduct.

MIT License and attribution

This project is free and open source under the MIT License. You may use, copy, modify, merge, publish, distribute, sublicense, and sell copies, including in commercial and closed-source products, provided the MIT copyright and permission notice is retained in copies or substantial portions.

Visible backlink credit is not required by MIT, but it is warmly appreciated:

• Creator: Siamese Cat Dev at DJAI Academy
• Original tools: Free PDF, Image, OCR, DOCX, QR, AI and Data Tools
• Community business: Siamese Cat Cafe

See ATTRIBUTION.md for copy-ready examples. The MIT License does not grant trademark rights or permission to imply endorsement; see TRADEMARKS.md.