Skip to content

README.md

Latest commit

 

History

History
186 lines (132 loc) · 6.81 KB

README.md

File metadata and controls

186 lines (132 loc) · 6.81 KB

AutosarFactory MCP Server

The MCP server exposes the autosarfactory API to AI coding assistants such as Claude Code. It allows an AI agent to query the full API reference, discover element hierarchies, look up enum values, and generate correct autosarfactory Python scripts — all without needing to read source code manually.

Setup

Install the required dependency first:

pip install mcp[cli]

Claude Code (quickest)

Run once from the repo root — Claude Code registers the server globally for this project:

claude mcp add autosarfactory -e PYTHONPATH=./mcp_server -- python -m autosarfactory_mcp.server

To include the optional --include-common-methods flag:

claude mcp add autosarfactory -e PYTHONPATH=./mcp_server -- python -m autosarfactory_mcp.server --include-common-methods

Verify the server is registered:

claude mcp list

VS Code / other MCP clients

There are two ways to run the server: stdio (recommended, VS Code manages the process) or HTTP (you start the server manually and VS Code connects to it).

Option A — stdio (recommended)

VS Code starts and stops the server automatically. No manual process management needed.

Add to .vscode/mcp.json:

{
  "servers": {
    "autosarfactory": {
      "type": "stdio",
      "command": "python",
      "args": ["-m", "autosarfactory_mcp.server"],
      "env": {
        "PYTHONPATH": "${workspaceFolder}/mcp_server"
      }
    }
  }
}

Server flags

Flag Default Description
--include-common-methods off Include arxml infrastructure base-class methods (set_timestamp, set_checksum, new_AdminData, new_Desc, new_Introduction, new_LongName, new_ShortNameFragment, new_Annotation) in get_class() responses. Omitted by default to keep responses lean — these methods are rarely needed for modelling.

To enable, add the flag to the args array:

"args": ["-m", "autosarfactory_mcp.server", "--include-common-methods"]

Option B — HTTP (manual server start)

Use this if you want to keep the server running persistently, share it across workspaces, or inspect its output directly.

1. Start the server manually (choose one transport):

# SSE transport (default port 8080)
cd mcp_server
PYTHONPATH=. python -m autosarfactory_mcp.server --transport sse --port 8080

# Streamable HTTP transport
cd mcp_server
PYTHONPATH=. python -m autosarfactory_mcp.server --transport streamable-http --port 8080

The --host flag defaults to 0.0.0.0. To bind to localhost only:

python -m autosarfactory_mcp.server --transport sse --host 127.0.0.1 --port 8080

2. Point .vscode/mcp.json at the running server:

{
  "servers": {
    "autosarfactory": {
      "type": "sse",
      "url": "http://localhost:8080/sse"
    }
  }
}

For streamable-http transport, use:

{
  "servers": {
    "autosarfactory": {
      "type": "http",
      "url": "http://localhost:8080/mcp"
    }
  }
}

What it does

  • find_creation_chain(elementType) — full creation chain from ArPackage down to an element in one call (use this first)
  • find_creators_of(elementType) — immediate parent class and method for a given element type
  • get_class(className, section) — cross-reference and attribute methods on a class (section: references, attributes, or all)
  • search_classes(keyword) — finds class names matching a keyword
  • get_enum(enumName) / list_enums() — valid enum values as ready-to-use Python expressions
  • get_system_element_map(useCase) — AUTOSAR element hierarchy for a modelling use case (SWC, CAN, Ethernet, DEXT etc.)
  • new_file_script_template() — boilerplate for creating a new arxml file
  • read_file_script_template() — boilerplate for reading and modifying existing arxml files
  • search_autosar_knowledge(query) — semantic search over AUTOSAR spec documents (requires a knowledge base, see below)
  • list_ecuc_modules() / get_ecuc_module() / get_ecuc_container() — ECUC/BSW module configuration (requires ecuc_param_def.json, see below)

Optional: AUTOSAR specification knowledge base

The search_autosar_knowledge tool becomes available when a pre-built knowledge base (.pkl file) is placed in mcp_server/autosarfactory_mcp/kb/. To build one from AUTOSAR spec PDFs or text files:

pip install sentence-transformers numpy pdfplumber
python mcp_server/kb_builder/build_knowledge_base.py --docs /path/to/autosar/specs/ --output autosar_kb.pkl

Multiple .pkl files are supported — the server merges them all at startup. Use distinct output names for different topic areas (e.g. autosar_com_kb.pkl, autosar_dext_kb.pkl).

Optional: ECUC parameter definition

The list_ecuc_modules, get_ecuc_module, and get_ecuc_container tools become available when a pre-built ecuc_param_def.json file is placed in mcp_server/autosarfactory_mcp/db/. This file captures the module → container → parameter / reference hierarchy from AUTOSAR ECUC module definition arxml files and is used by the agent to look up container structure and value classes when generating ECU configuration scripts.

Use the ecuc_module_def_to_json.py script (at the repo root) to generate or update this file from your own arxml files.

Generate from scratch

python ecuc_module_def_to_json.py <file.arxml> [<file2.arxml> ...] -o mcp_server/autosarfactory_mcp/db/ecuc_param_def.json

Update specific modules (merge mode)

Modules found in the arxml replace their counterpart in the existing JSON by name. Modules not present in the arxml are left untouched. New modules are appended.

# Update in-place
python ecuc_module_def_to_json.py <file.arxml> --update mcp_server/autosarfactory_mcp/db/ecuc_param_def.json

# Update only a specific module
python ecuc_module_def_to_json.py <file.arxml> --module CanIf --update mcp_server/autosarfactory_mcp/db/ecuc_param_def.json

# Update and write to a different file
python ecuc_module_def_to_json.py <file.arxml> --update mcp_server/autosarfactory_mcp/db/ecuc_param_def.json -o ecuc_param_def_new.json

After updating the file, restart the MCP server so it reloads the new data.

AUTOSAR element map

The file autosarfactory_mcp/ar_map/autosar_element_map.md contains a human-readable and agent-readable map of which AUTOSAR elements are needed for each modelling use case (Sender-Receiver, CAN, Ethernet/SOME/IP, DEXT, calibration, etc.). This file is loaded automatically at server startup and can be extended without modifying any Python code — just edit the .md file and restart the MCP server.