MCP server for B&R Automation Studio help documentation search. Provides keyword search by default using LanceDB's native full-text search (FTS), and optional hybrid semantic + keyword search using Reciprocal Rank Fusion (RRF) when an embedding API is configured.

• Keyword search (default): Fast full-text search using LanceDB's native FTS — no external dependencies
• Hybrid search (optional): RRF fusion of vector similarity and keyword matching when embeddings are enabled
• API-based embeddings: Works with any OpenAI-compatible endpoint (Ollama, OpenAI, Azure OpenAI, GitHub Models, LiteLLM) — no local ML models required
• Smart ranking: Query-type detection shifts weights between FTS and vectors (identifiers like MC_MoveAbsolute favor exact match; natural language favors semantic similarity)
• Category filtering and hierarchical browsing
• Auto-generated links to B&R online help (AS4/AS6)
• HelpID lookup for context-sensitive help integration
• Incremental reindexing — only changed pages are re-processed
• Two-phase build: keyword search available within minutes while embeddings build in the background

• B&R Automation Studio installed (with help documentation)
• VS Code with GitHub Copilot extension
• For standalone binary: Download as-help-server.exe from Releases — no Python or Docker required
• For UV: Python 3.12+ with uv
• For Docker: Docker Desktop
• Optional (for hybrid search): An OpenAI-compatible embedding API (e.g., Ollama with nomic-embed-text)

Add to .vscode/mcp.json in your workspace:

No Python, no Docker — just download the .exe from Releases and place it in %APPDATA%\as-help-mcp\.

{
  "servers": {
    "as-help": {
      "command": "${env:APPDATA}\\as-help-mcp\\as-help-server.exe",
      "args": [
        "--help-root",
        "C:\\Program Files (x86)\\BRAutomation\\AS6\\Help-en\\Data",
        "--db-path",
        "${env:APPDATA}\\as-help-mcp\\data\\as6\\.ashelp_lance",
        "--metadata-dir",
        "${env:APPDATA}\\as-help-mcp\\data\\as6\\.ashelp_metadata",
        "--as-version",
        "6"
      ]
    }
  }
}

Update --help-root to match your AS installation:

• AS 4.x: C:\\BRAutomation\\AS412\\Help-en\\Data
• AS 6.x: C:\\Program Files (x86)\\BRAutomation\\AS6\\Help-en\\Data

{
  "servers": {
    "as-help": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-v", "C:\\Program Files (x86)\\BRAutomation\\AS6\\Help-en\\Data:/data/help:ro",
        "-v", "ashelp-data:/data/db",
        "-e", "AS_HELP_VERSION=6",
        "-e", "AS_HELP_FORCE_REBUILD=false",
        "ghcr.io/brdk-public/as-help-mcp:latest"
      ]
    }
  }
}

Update the volume path to match your AS installation:

• AS 4.x: C:\\BRAutomation\\AS412\\Help-en\\Data:/data/help:ro
• AS 6.x: C:\\Program Files (x86)\\BRAutomation\\AS6\\Help-en\\Data:/data/help:ro
• AS 6.x in WSL: /mnt/c/Program Files (x86)/BRAutomation/AS6/Help-en/Data:/data/help:ro

{
  "servers": {
    "as-help": {
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "/path/to/as-help-mcp",
        "as-help-server",
        "--help-root",
        "C:\\Program Files (x86)\\BRAutomation\\AS6\\Help-en\\Data",
        "--db-path",
        "..\\data\\as6\\.ashelp_lance",
        "--metadata-dir",
        "..\\data\\as6\\.ashelp_metadata",
        "--as-version",
        "6"
      ]
    }
  }
}

Update --directory to point to your cloned repository and adjust paths as needed.

Restart VS Code, then test in Copilot Chat: "Search AS help for mapp Motion"

First run takes 2-3 minutes to build the keyword search index. With embeddings enabled, a full hybrid build takes 15-20 minutes (keyword search is available immediately while embeddings build in the background). Subsequent starts are instant (~3s).

By default, the server uses keyword-only search (FTS). To enable hybrid semantic + keyword search, configure an OpenAI-compatible embedding API.

1. Install Ollama and pull an embedding model:

ollama pull nomic-embed-text

2. Add --create-embeddings true and embedding environment variables to your MCP config:

{
  "servers": {
    "as-help": {
      "command": "${env:APPDATA}\\as-help-mcp\\as-help-server.exe",
      "args": [
        "--help-root", "C:\\Program Files (x86)\\BRAutomation\\AS6\\Help-en\\Data",
        "--db-path", "${env:APPDATA}\\as-help-mcp\\data\\as6\\.ashelp_lance",
        "--metadata-dir", "${env:APPDATA}\\as-help-mcp\\data\\as6\\.ashelp_metadata",
        "--as-version", "6",
        "--create-embeddings", "true"
      ],
      "env": {
        "EMBEDDING_API_ENDPOINT": "http://localhost:11434",
        "EMBEDDING_API_KEY": "ollama",
        "EMBEDDING_MODEL": "nomic-embed-text",
        "EMBEDDING_DIMENSIONS": "768",
        "EMBEDDING_BATCH_SIZE": "100",
        "EMBEDDING_MAX_CHARS": "4000"
      }
    }
  }
}

Any OpenAI-compatible endpoint works — OpenAI, Azure OpenAI, GitHub Models, LiteLLM, etc. Just update the endpoint, key, model, and dimensions accordingly.

When embeddings are enabled, the server uses Reciprocal Rank Fusion (RRF) to combine four search signals:

Query-type detection automatically selects weights: identifier queries (e.g., MC_MoveAbsolute, X20DI9371) shift toward FTS + title match; natural language queries favor vector similarity.

For a deep dive into the RAG architecture — chunking strategy, two-phase build, RRF fusion, embedding model choice, and alternatives considered — see RAG.md.

# Clone and install
git clone <repository-url>
cd as-help-mcp
uv sync --extra test --extra dev

# Run server with command line arguments (precedence over .env)
uv run as-help-server --help-root "C:\BRAutomation\AS412\Help-en\Data" --db-path "data\.ashelp_lance" --metadata-dir "data\.ashelp_metadata"

# Or use relative paths (automatically resolved)
uv run as-help-server --db-path ./data/lance_index --metadata-dir ./data

You can also create a .env file in the root directory. Command-line arguments will override these values if provided.

AS_HELP_ROOT=C:\Program Files (x86)\BRAutomation\AS6\Help-en\Data
AS_HELP_VERSION=6

Run uv run as-help-server --help for full details.

These are only needed when --create-embeddings true is set:

# Local build
docker compose build

# Run with your help files mounted
docker compose run --rm \
  -v "C:\Program Files (x86)\BRAutomation\AS6\Help-en\Data:/data/help:ro" \
  as-help-local

The MCP Inspector provides a web UI for testing tools and prompts:

# With UV
uv run mcp dev src/server.py

# Opens browser at http://localhost:5173

Note: On Windows, use VS Code's Run and Debug panel instead (stdio transport issues with Inspector).

Use the launch configurations in .vscode/launch.json:

1. Rebuild BR Help Index - First run to build index
2. Run BR Help MCP Server - Normal server startup
3. Test BR Help Indexer - Quick XML parse test

{
  "servers": {
    "as-help-4": {
      "command": "${env:APPDATA}\\as-help-mcp\\as-help-server.exe",
      "args": [
        "--help-root", "C:\\BRAutomation\\AS412\\Help-en\\Data",
        "--db-path", "${env:APPDATA}\\as-help-mcp\\data\\as4\\.ashelp_lance",
        "--metadata-dir", "${env:APPDATA}\\as-help-mcp\\data\\as4\\.ashelp_metadata",
        "--as-version", "4"
      ]
    },
    "as-help-6": {
      "command": "${env:APPDATA}\\as-help-mcp\\as-help-server.exe",
      "args": [
        "--help-root", "C:\\Program Files (x86)\\BRAutomation\\AS6\\Help-en\\Data",
        "--db-path", "${env:APPDATA}\\as-help-mcp\\data\\as6\\.ashelp_lance",
        "--metadata-dir", "${env:APPDATA}\\as-help-mcp\\data\\as6\\.ashelp_metadata",
        "--as-version", "6"
      ]
    }
  }
}