Allows the agent to perform web searches using DuckDuckGo as a fallback search engine, retrieving repair guides and forum discussions to aid automotive diagnostics.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Car_Diagnostic_AI Diagnose why my car goes into limp mode"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Car_Diagnostic_AI — also answering to CDA and Majster-AI — is a diagnostic agent that connects a large language model to a car's control modules over UDS/CAN, to a local index of workshop manuals, and to live web search. The three capabilities are exposed as independent Model Context Protocol servers and orchestrated by a LangGraph agent with a hard human-in-the-loop gate on anything that writes to the vehicle.
Developed and tested against a Land Rover Freelander 2 (2010, 2.2 TD4).
It ships with a Cyber-HUD web interface: an interactive 3D chassis whose module pins sit at real vehicle coordinates, live telemetry gauges, and a drag-to-authorize gate for anything that writes to the car.
⚠️ It defaults to a simulator, and to READ_ONLY
A fresh clone cannot transmit on a real bus and cannot write to a vehicle. Both take a deliberate configuration change. Read docs/SAFETY.md before connecting to a car you care about.
you> The car goes into limp mode on the motorway. What's wrong with it?
[agent.plan] Calling: read_dtc
[car.read_dtc] Reading DTCs from ECM
[agent.plan] Calling: read_live_data
[agent.plan] Calling: search_manual
majster> Three codes stored in the ECM, and I think two of them are one fault:
P0299 (confirmed) — turbocharger underboost
P2015 (confirmed) — intake manifold runner position sensor
P0401 (pending) — EGR flow insufficient
Live data backs up the underboost: manifold pressure reads 101 kPa at idle
against 100 kPa barometric, so the turbo isn't building any boost at all.
P0299 and P2015 together on a DW12 is a familiar pattern — both are air-path
faults and the swirl-flap linkage sits right next to the actuator. The
manual (fl2_engine.pdf, page 412) says actuator vacuum should reach 0.6 bar
at idle; that's the measurement that would separate a seized actuator rod
from a boost leak, and it's a five-minute check before anyone buys a turbo.
The pending EGR code I'd leave alone for now — it hasn't confirmed, and a
fouled EGR is a plausible consequence of running underboost rather than a
separate fault.Note what it did not do: guess, and offer to clear the codes.
Related MCP server: Codex MCP Server
+--------------------------------------------------+
| Cyber-HUD (React · Three.js · Framer Motion) |
+---------------------------+----------------------+
| WS /ws/diagnostics
+---------------------------+----------------------+
| FastAPI (telemetry · agent · HITL) |
+---------------------------+----------------------+
|
+--------------+------------+
| LangGraph agent |
| Claude Opus 5 / Ollama |
+-------------+-------------+
|
+-----------------+-----------------+
| | |
+---------v------+ +--------v-------+ +-------v--------+
| Car_Interface | | RAG_Workshop | | Web_Search |
| MCP | | MCP | | MCP |
+---------+------+ +--------+-------+ +-------+--------+
| | |
UDS over CAN local manuals Tavily / DuckDuckGoCar_Interface_MCP — python-can + udsoncan. Reads DTCs, live data and
raw DIDs from any module; clears codes only through the approval handshake.
Five interchangeable backends plus a built-in ECU simulator.
RAG_Workshop_MCP — ChromaDB over your own workshop manual PDFs. Runs entirely on-device; every answer carries a file-and-page citation.
Web_Search_MCP — Tavily with a keyless DuckDuckGo fallback, weighted towards Land Rover forums.
The orchestrator — READ_ONLY by default; every write pauses the graph and waits for a human.
Full detail in docs/ARCHITECTURE.md.
git clone https://github.com/Mati83mon/Car_Diagnostic_Ai.git
cd Car_Diagnostic_Ai
python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -e ".[all,dev]" # or: pip install -r requirements.txt
cp .env.example .env # then edit itOn Termux or another ARM device where the heavy extras will not build, skip them — the project is designed to work without:
pip install -e ".[car,mcp,agent,web]"majster-ai doctorAlso available as cda and car-diagnostic-ai, or python main.py <command>.
Everything lives in .env. The defaults are safe, so an empty file is a valid
file. The settings that matter most:
# --- safety -----------------------------------------------------------
MAJSTER_WRITE_ENABLED=false # master switch. Leave false.
MAJSTER_REQUIRE_APPROVAL=true # human-in-the-loop. Leave true.
# --- vehicle interface -------------------------------------------------
MAJSTER_CAN_BACKEND=virtual # virtual|socketcan|slcan|serial|j2534|rfcomm
MAJSTER_CAN_CHANNEL=can0
MAJSTER_CAN_BITRATE=500000
# --- LLM ----------------------------------------------------------------
ANTHROPIC_API_KEY=sk-ant-... # Claude Opus 5; falls back to Ollama if unset
MAJSTER_OLLAMA_MODEL=qwen2.5:7b-instruct
# --- web search ---------------------------------------------------------
TAVILY_API_KEY=tvly-... # optional; DuckDuckGo is used without itSee .env.example for every option, annotated.
majster-ai chatmajster-ai ask "why is the DPF light on?"majster-ai dtc --module ECM # read fault codes
majster-ai dtc --all # scan every module
majster-ai live RPM COOLANT_TEMP MAF # read live data
majster-ai scan # discover which ECUs answer
majster-ai clear --module ECM # write: prompts for approvalcp ~/manuals/*.pdf data/manuals/
majster-ai ingest
majster-ai search "swirl flap removal procedure"Manuals are indexed and searched locally. Nothing is uploaded.
cd frontend && npm install && npm run build # once
majster-ai web # http://127.0.0.1:8000For frontend development, run the Vite dev server alongside it:
majster-ai web # terminal 1 — API on :8000
cd frontend && npm run dev # terminal 2 — UI on :5173, proxying to :8000The UI streams live telemetry, shows module health on a 3D chassis, and pauses
for a drag-to-authorize gesture before any write. Clicking a fault code flies
the camera to the component it refers to — a chassis code like C0034 goes to
the front-right wheel sensor, not to the ABS module in the engine bay.
majster-ai web binds to 127.0.0.1 by default. Anything that can reach the
port can ask the agent to propose a write; the approval gate still holds, but
the prompt would be answered by whoever is there. Only use --host 0.0.0.0 on
a network you control.
majster-ai serve car_interface # stdio transportConfiguration for Claude Desktop and friends is in docs/ARCHITECTURE.md.
Four independent layers stand between the model and the vehicle. A write has to get through all of them.
Layer | Guarantee |
1. Master switch |
|
2. Token handshake | The first call always fails and returns an impact summary plus a single-use token, bound to a hash of the exact arguments and expiring in five minutes. Approval for the ECM cannot clear the ABS module. |
3. Human pause | The graph suspends via |
4. System prompt | Tells the model what the other three layers will do. Treated as the weakest layer, because it is. |
The browser is simply another approver. The slider sends one boolean; the confirmation token is created and redeemed inside the server process and never appears in any WebSocket frame. A client can answer the question the server chose to ask — it can never pose one, and it can never mint the credential that performs the write.
Layer 2 lives in the service behind the MCP server, so it protects the car even when something other than this agent is driving.
========================================================================
WRITE OPERATION - HUMAN APPROVAL REQUIRED
========================================================================
Operation : clear_dtc
Module : ECM (Engine Control Module - 2.2 TD4)
Scope : ALL stored DTCs in this module
Risk : MEDIUM Reversible: NO
Will erase 3 code(s):
- P0299-00 Turbocharger/Supercharger A Underboost Condition
- P2015-00 Intake Manifold Runner Position Sensor/Switch Circuit
- P0401-00 Exhaust Gas Recirculation Flow Insufficient Detected
Consequences:
! Freeze-frame data captured when the fault occurred will be lost.
! Readiness monitors reset; the vehicle may fail an emissions test.
! Clearing does not repair anything. If the fault is still present
the code will return.
========================================================================
Type 'yes' to authorise, anything else to decline:Full detail in docs/SAFETY.md.
Backend | Interface | Platform |
| none — built-in simulator | anywhere |
| Tactrix Openport 2.0 | Linux, Windows |
| USB2CAN, CANable, PiCAN | Linux |
| CANable/CANtact (slcan firmware) | Linux, macOS, Windows |
| ELM327 / OBDLink over Bluetooth | Linux, Termux |
The Freelander 2 uses ISO 15765-4 CAN at 500 kbit/s with 11-bit identifiers, on OBD-II pins 6 (CAN-H) and 14 (CAN-L).
Setup for each, plus Termux, Raspberry Pi, and which ELM327 adapters actually work: docs/HARDWARE.md.
Only two diagnostic addresses on any car are legislated and therefore certain:
the powertrain addresses 0x7E0/0x7E8 and 0x7E1/0x7E9. Everything else
in the built-in module map is community-derived and ships marked
verified: false.
The same applies to live-data scaling: the SAE J1979 PIDs are standard and marked verified; manufacturer DIDs are proprietary and none ship at all, because a confidently-wrong number is the worst possible output from a diagnostic tool.
To find out what is true for your car:
majster-ai scanThen record what answered in data/modules.json. See
docs/FREELANDER2.md.
make install-dev
make check # black --check, flake8, pytest
make test-cov # coverage reportThe entire suite runs against the in-process ECU simulator — no hardware, no API key, no network:
736 passed, 1 skippedThe simulator is a real UDS implementation rather than a mock, so the retry logic, the DTC codec, the MCP tools and the HITL gate all run against the same byte stream they will see on a real bus. Fault injection makes the flaky-bus paths deterministic:
ecm.inject_faults(drop_next=2) # two silent timeouts, then fine
ecm.inject_faults(pending_next=3) # three NRC 0x78, then the answer
ecm.inject_faults(busy_next=1) # one NRC 0x21 busyRepeatRequestCI runs lint, the suite on Python 3.10/3.11/3.12, the safety invariants as their own job, and an integration job that spawns the MCP servers as real subprocesses.
majster_ai/
├── agent/ LangGraph orchestrator, HITL, LLM providers
├── mcp_servers/
│ ├── car_interface/ UDS/CAN — the safety gate lives in service.py
│ ├── rag_workshop/ local manual retrieval
│ └── web_search/ Tavily / DuckDuckGo
├── web/ FastAPI + /ws/diagnostics, WebSocketApprover
├── config.py settings and the two safety gates
└── cli.py
frontend/ React + Three.js Cyber-HUD (see frontend/README.md)
tests/ 736 tests, all hardware-free
docs/ ARCHITECTURE, SAFETY, HARDWARE, FREELANDER2This project is for educational and research purposes. It is not a certified diagnostic tool and is not a substitute for the manufacturer's equipment or a qualified technician.
The authors accept no liability for damage to a vehicle, damage to control modules, failed repairs, voided warranties, or injury arising from use of this software. You are responsible for understanding what any command does before it reaches your car's bus. If you are not certain, do not send it.
Projekt służy wyłącznie do celów edukacyjnych i badawczych. Autor nie ponosi odpowiedzialności za jakiekolwiek uszkodzenia pojazdu, uszkodzenia sterowników (ECU) lub obrażenia ciała wynikające z użytkowania tego oprogramowania.
Zawsze upewnij się, że wiesz, jakie komendy — zwłaszcza polecenia ZAPISU — są wysyłane na magistralę CAN Twojego samochodu.
MIT — see LICENSE.
Workshop manuals are copyrighted and are not distributed with this project.
data/manuals/ is gitignored; source your own legally.
This server cannot be installed
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.
Let AI operate servers without SSH. Choose actions, approve risky changes, and audit every step.
Sovereign Agent OS — Persistent Memory, Governance & Compliance for AI Agents.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/Mati83mon/Car_Diagnostic_Ai'
If you have feedback or need assistance with the MCP directory API, please join our Discord server