Skip to content

Releases: jbulger82/Fullstack_MCP_hub

v0.1.0 – First public Alpha

15 Dec 19:30
d3c6df1

Choose a tag to compare

🚀 FullStack MCP Hub v0.1.0 – First Public Alpha

This is the first public release of FullStack MCP Hub — a complete MCP toolkit with:

  • a central MCP gateway + hub
  • a clean browser UI
  • built-in RAG system
  • and 50+ free tools (no paid APIs required)

It runs locally and can be plugged into almost any modern LLM client.


🌐 Compatibility

This release is designed to work with:

  • OpenAI / ChatGPT Codex CLI
  • Gemini CLI
  • ChatGPT (Dev Mode) via custom MCP connector
  • Claude desktop / other MCP-aware clients
  • Local LLMs (llama.cpp, LM Studio, etc. — anything that can use an MCP endpoint)

If your client can talk MCP over stdio or SSE, you can point it at this hub and instantly reuse the same tool + RAG stack.


✨ Highlights in v0.1.0

1. Full MCP gateway + hub

  • Central hub that connects to multiple MCP servers (stdio and SSE).
  • Normalizes tools from all servers into a single list.
  • Applies description overrides and a persistent blocklist.
  • Exposes:
    • GET /tools – list all tools
    • GET /sse – universal MCP endpoint
    • POST /gemini/v1/execute – Gemini-style adapter

2. Browser UI (GUI)

A React/Vite UI served by the gateway:

  • Servers view

    • Add stdio or SSE servers via a guided form.
    • Test connection before saving.
    • Persist config into tool-registry/master.json.
  • Tools view

    • Browse tools from all connected servers.
    • See description + input schema.
    • Run tools with JSON payloads.
    • Save & reuse per-tool presets.
    • Block tools you don’t want used (saved to tool-blocklist.json).
  • Blocked view

    • See blocked tools.
    • Restore tools back into the main list.
  • RAG view

    • Drag & drop files into the RAG storage.
    • Browse uploads, saved_chats, indexes, images, and profile directories.
    • Trigger index creation and search from the UI.

The goal is to make MCP feel like a tool control panel, not just a JSON spec.


📚 Built-in RAG (“Generation 1.5”)

This release ships a local RAG system that does chunked + fuzzy search without embeddings:

  • local_rag__create_index

    • Scans a directory under data/rag/.
    • Splits text into ~500-word chunks with overlap.
    • Stores the chunks and metadata in a persisted indexes.pkl.
  • local_rag__search_index

    • Keyword + fuzzy search across chunks.
    • Supports filters:
      • path_contains
      • tag (from #tags: lines in files)
      • modified time (mtime) ranges.
    • Returns matching chunks with file + chunk metadata.
  • local_rag__save_chat

    • Saves raw + summary chat transcripts into data/rag/saved_chats/.
    • Never overwrites: filenames include timestamp + model.
  • local_rag__save_image

    • Saves base64 image data into data/rag/images/.
  • local_rag__list_files, local_rag__list_indexes, local_rag__read_file

    • Introspection helpers to see what’s in RAG and inspect content.

This sits between simple keyword search and full vector RAG: usable today with no separate embedding model, and easy to extend to vectors later.


🧰 Bundled MCP servers & tools (no paid APIs)

This release includes a solid default tool stack:

Filesystem

Full read/write + introspection within allowed roots:

  • filesystem__create_directory
  • filesystem__directory_tree
  • filesystem__edit_file
  • filesystem__get_file_info
  • filesystem__list_allowed_directories
  • filesystem__list_directory
  • filesystem__list_directory_with_sizes
  • filesystem__move_file
  • filesystem__read_file
  • filesystem__read_media_file
  • filesystem__read_multiple_files
  • filesystem__read_text_file
  • filesystem__search_files
  • filesystem__write_file

Local RAG

  • local_rag__create_index
  • local_rag__search_index
  • local_rag__list_files
  • local_rag__list_indexes
  • local_rag__read_file
  • local_rag__save_chat
  • local_rag__save_image

Shell

  • shell__run_command

    Powerful: lets the model execute commands on your machine. Intended for local / trusted use.

Playwright (browser automation)

  • playwright__browser_navigate
  • playwright__browser_screenshot
  • playwright__browser_click
  • playwright__browser_click_text
  • playwright__browser_fill
  • playwright__browser_select
  • playwright__browser_select_text
  • playwright__browser_hover
  • playwright__browser_hover_text
  • playwright__browser_evaluate

Python REPL (with its own venv)

  • python_repl__exec
  • python_repl__reset
  • python_repl__pip_install

SQLite

  • sqlite__read_query
  • sqlite__write_query
  • sqlite__create_table
  • sqlite__list_tables
  • sqlite__describe_table
  • sqlite__append_insight

Web search (basic & advanced)

  • websearch__web_search (DuckDuckGo)
  • websearch_adv__full-web-search (multi-engine + extraction)
  • websearch_adv__get-single-web-page-content (robust single-page scraper)

Research

  • research__wikipedia_search
  • research__arxiv_search
  • research__images_search_commons

Scraper

  • scrape__scrape_page (HTML → cleaned text + title)

Image generation

  • pollinations__generateImageUrl
  • pollinations__listImageModels

CoinGecko (SSE-based)

  • coingecko__get_simple_price
  • coingecko__get_coins_markets
  • coingecko__get_range_coins_market_chart
  • coingecko__get_search

Clock

  • clock__now
  • clock__add_delta

Every tool in this release has been smoke-tested by hand end-to-end.


🔧 Getting started (summary)

For detailed install instructions, see the main README, but the basics are:

git clone https://github.com/<your-username>/FullStack_MCP_Hub.git
cd FullStack_MCP_Hub
export MCP_ROOT="$(pwd)"
chmod +x setup.sh start.sh
./setup.sh
./start.sh