AaronWiseAI MCP Ecosystem
AaronWiseAI System map
LM Studio + MCP · snapshot of the 2026-10-22 transfer

LM Studio MCP Ecosystem

Every MCP server LM Studio loads, what each one talks to, and where its files end up. Built from your mcp.json, the transfer README and SYSTEM_GRAPH.md. API keys, tokens and passwords are left out.

  • 21 MCP servers
  • 19 local (stdio)
  • 2 over HTTP
  • 12 Python, one shared venv
  • 7 Node (2 direct, 5 npx)

Items tagged inferred come from file names or earlier diagrams. The code behind them was not read.

The whole system

LM Studio reads one file, mcp.json, and starts or connects to 21 servers. Those servers reach outside services and write into a few folders on this PC.

You and the client
Youchat inside LM Studio
LM Studiodesktop app, acts as the MCP client
mcp.jsonbackup in lmstudio-config\ · live copy at C:\Users\awise\.lmstudio\mcp.json
Memory · 3
memory
long_term_memory
AI-Vault
Automation · 4
task-engine
terminal-bridge
maton-gateway
lmstudio-stack
Publish & comms · 4
wordpress
aaronwiseai-sftp
aaronwiseai-business-email
obsidian
Local AI & media · 5
local-tts
lemonade
computer-vision
application-intelligence
alibaba-model-studio
Files · 1
filesystemTransfer folder, Desktop, Documents, Downloads
Outside services and local apps
DuckDuckGo · Bing · Braveweb-search
Live websitesbrowser (Playwright)
Context7 APIcontext7
Hugging Facelong_term_memory (embedding model), huggingface-explicit-only
Maton APImaton-gateway
Alibaba DashScope (intl)alibaba-model-studio
IONOS webspaceaaronwiseai-sftp
IONOS mailaaronwiseai-business-email
aaronwiseai.com/learnaiwordpress
Obsidian :27124obsidian
Lemonade 192.168.12.174:13305lemonade (another machine on the LAN)
AI-Vault localhost:3002AI-Vault
Folders on this PC · root is Desktop\LMStudio-MCP-Transfer
AI Output\alibaba, sftp, email, task-engine
AI-Memory\ChromaDB, task-engine data
LM-Studio-Files\audio, images, models, task markdown
Maton-MCP\scripts, venv, logs, journals
lmstudio-config\mcp.json, server-logs
application-skills\ · computer-vision\read by application-intelligence and computer-vision
Colors mark the six server groups. They stay the same in every diagram below.

All 21 servers

One card per entry in mcp.json. Filter by group.

Showing 21 of 21

memory

stdionpx

Stock knowledge-graph memory. No settings in mcp.json.

Runs
@modelcontextprotocol/server-memory
Memory

long_term_memory

stdioPython

Vector memory in ChromaDB. Text is embedded with nomic-embed-text-v1.5 and recalled by similarity.

Script
long-term-memory-mcp\long_term_memory_mcp.py
Data
AI-Memory\long-term-memory\data
Reaches
Hugging Face (model download)
Flags
integrity check off · auto-rebuild off
Memory

AI-Vault

HTTP

Local vault server. You start it yourself; LM Studio only connects.

URL
http://localhost:3002/mcp
Auth
bearer token
Check
ai_vault_mcp_check.mjs (root)
Memory

task-engine

stdioPython

Task queue you can hand work to. Details in the Task engine section.

Script
Maton-MCP\task_engine_mcp.py
Data
AI-Memory\task-engine\data
Output
AI Output\
Polling
every 900 s
Automation

terminal-bridge

stdioPython

Runs scripts and commands inside allowed folders. Plain CMD access is switched off.

Script
Maton-MCP\terminal_bridge_mcp.py
Allowed
Maton-MCP, LM-Studio-Files, Desktop, Documents, Downloads
CMD
disabled (0)
Automation

maton-gateway

stdioPython

Gateway to Maton services in router mode.

Script
Maton-MCP\maton_mcp_server.py
Reaches
Maton API (key set)
Timeout
30 s
Automation

lmstudio-stack

stdioPythoninferred

Looks at the state of the LM Studio stack. Purpose taken from the file name.

Script
Maton-MCP\lmstudio_stack_mcp.py
Automation

wordpress

stdioNode

Reads and edits the WordPress site over its MCP endpoint.

Runs
wordpress-mcp-runtime\...\mcp-wordpress-remote\dist\proxy.js
Site
aaronwiseai.com/learnai
Auth
application password · OAuth off
Publish & comms

aaronwiseai-sftp

stdioPython

Uploads files to the IONOS webspace. Files land in aidocs\ and are public on the site.

Script
Maton-MCP\ftp_sftp_mcp.py
Port
22 (SFTP)
Public URL
aaronwiseai.com/aidocs/
Local copy
AI Output\
Publish & comms

aaronwiseai-business-email

stdioPython

Reads and sends business email through IONOS.

Script
Maton-MCP\ionos_mail_mcp.py
IMAP
imap.ionos.com : 993
SMTP
smtp.ionos.com : 587
Local copy
AI Output\
Publish & comms

obsidian

stdionpx

Reads and writes notes in the Obsidian vault.

Runs
obsidian-mcp-server
API
https://127.0.0.1:27124
Vault
AI-Vault
Publish & comms

local-tts

stdioPython

Text to speech with Piper, fully local.

Script
Maton-MCP\local_tts_mcp.py
Output
LM-Studio-Files\audio\tts
Voices
...\audio\tts\voices
Engine
...\tts\piper\piper\piper.exe
Local AI & media

lemonade

stdioPython

Sends image, audio and 3D-mesh jobs to a Lemonade server on another machine and saves what comes back.

Script
Maton-MCP\lemonade_mcp.py
Server
192.168.12.174:13305
Timeouts
mesh and image 1800 s · default wait 45 s
Inline images
up to 12 MB
Local AI & media

computer-vision

stdioPythoninferred

Image and screen understanding. Purpose taken from the name.

Script
computer-vision\computer_vision_mcp.py
Local AI & media

application-intelligence

stdioPython

Serves per-application skills from a folder.

Script
application_intelligence_mcp.py (root)
Skills root
application-skills\
Local AI & media

alibaba-model-studio

stdioPython

Routes requests to Alibaba Model Studio, limited to free models.

Script
Maton-MCP\alibaba_model_studio_router_mcp.py
Endpoint
dashscope-intl.aliyuncs.com (compatible mode)
Free only
on · timeout 60 s
Output
AI Output\
Local AI & media

filesystem

stdionpx

Read and write access to four folders.

Runs
@modelcontextprotocol/server-filesystem
Roots
LMStudio-MCP-Transfer, Desktop, Documents, Downloads
Files

How each server starts

Twelve servers share one Python environment. Two run on Node directly, five start through npx, and two are plain HTTP addresses.

mcp.json entry
command + args + envor a url + headers for HTTP
Python · 12 servers
Shared venvMaton-MCP\long-term-memory-mcp\.venv\Scripts\python.exe
terminal_bridge_mcp.py
task_engine_mcp.py
maton_mcp_server.py
lmstudio_stack_mcp.py
long_term_memory_mcp.py
ftp_sftp_mcp.py
ionos_mail_mcp.py
local_tts_mcp.py
lemonade_mcp.py
alibaba_model_studio_router_mcp.py
computer-vision\computer_vision_mcp.py
application_intelligence_mcp.pylives in the root folder
Node, direct · 2 servers
wordpressC:\Program Files\nodejs\node.exe running wordpress-mcp-runtime\...\proxy.js
HTTP · 2 servers
AI-Vaulthttp://localhost:3002/mcp
Most Python servers set PYTHONUTF8=1. If the venv breaks on a new PC, all 12 Python servers break together.

Memory and knowledge

Three different stores sit behind the memory tools. They do not share data with each other as far as the config shows.

Working graph

memory

Knowledge graph, stock npx server.

Long-term recall

long_term_memory

Python server saves and searches memories.

nomic-embed-text-v1.5

Model from Hugging Face. Remote code trusted. HF token set.

ChromaDB

AI-Memory\long-term-memory\data

Notes vault

obsidian

Talks to the Obsidian app on 127.0.0.1:27124, vault AI-Vault.

AI-Vault (HTTP)

Server at localhost:3002/mcp.

Also in the stack: memory_bridge_learning_memory.json (used by long_term_memory per SYSTEM_GRAPH) and lemonade_learning_memory.json (used by lemonade).
Worth checking. The obsidian server and the AI-Vault server share a name. Nothing in the config says whether port 3002 serves the same vault Obsidian opens. The README also lists ChromaDB under Maton-MCP\long-term-memory-mcp\data, while mcp.json points to AI-Memory\long-term-memory\data. Back up both until you know which one is live.

Task engine

LM Studio only launches task_engine_mcp.py. A supervisor and a worker script sit next to it in Maton-MCP\. The config does not show how those two are started.

Client
You in LM Studioask to create, list or check tasks
MCP server
task_engine_mcp.pythe only piece LM Studio launches
Queue and state
AI-Memory\task-engine\dataqueue, state, metadata (TASK_ENGINE_DATA_DIR)
LM-Studio-Files\task-engine\markdown_tasksreadable task files (TASK_ENGINE_TASK_QUEUE_OUTPUT_DIR)
Maton-MCP\mcp_task_journal.mdshared task journal
Maton-MCP\task_engine_dataolder or extra data folder listed in SYSTEM_GRAPH
Supervisor
task_engine_supervisor.pywatches the queue, schedules, spots stuck or failed tasks. Fed by the service registry (maton_service_registry_complete.py, _ext.py).
Worker
task_engine_worker.pyclaims a task, does the work, records the result.
Output
AI Output\TASK_ENGINE_OUTPUT_ROOT for task-engine
Both pollers use 900 seconds, which is 15 minutes. A new task can wait about that long before a worker looks at it, if the workers poll on that interval.

Life of a task inferred

  1. Create You

    Ask LM Studio for a task. task_engine_mcp.py writes it to the queue and a markdown file.

  2. Scan and schedule Supervisor

    Looks for pending tasks every 900 s and sets order.

  3. Claim Worker

    Polls every 900 s, takes one task, marks it in progress.

  4. Do the work Worker

    Runs the task and may call other MCP tools.

  5. Save results Worker

    Writes files to AI Output\ or sends them on.

  6. Close out Worker

    Marks the task done or failed and logs it to the journal.

  7. Watch Supervisor

    Checks for stuck or failed tasks.

  8. Check status You

    Ask for the task list or one task. The MCP server reads the queue and answers.

Servers a task may call inferred

  • web-search
  • browser
  • terminal-bridge
  • local-tts
  • long_term_memory
  • aaronwiseai-sftp
  • aaronwiseai-business-email
  • wordpress
  • obsidian
  • lemonade
  • computer-vision
  • application-intelligence
  • context7

Settings in mcp.json

TASK_ENGINE_DATA_DIR
AI-Memory\task-engine\data
TASK_ENGINE_OUTPUT_ROOT
AI Output
TASK_ENGINE_TASK_QUEUE_OUTPUT_DIR
LM-Studio-Files\task-engine\markdown_tasks
WORKER_POLL_INTERVAL_SECONDS
900
SUPERVISOR_POLL_INTERVAL_SECONDS
900
TASK_ENGINE_LOG_STREAM
stdout
Mismatch to know about. local-tts also sets TASK_ENGINE_OUTPUT_ROOT, but to LM-Studio-Files instead of AI Output.

Publishing and communications

Four servers move content out to your site, your mailbox and your notes. SFTP is the only one that makes files public, under aaronwiseai.com/aidocs/.

wordpress

Node proxy, app password

aaronwiseai.com/learnai

WordPress site

aaronwiseai-sftp

Python server

IONOS webspace, aidocs\

Public at aaronwiseai.com/aidocs/

aaronwiseai-business-email

Python server

IONOS mail

business mailbox

obsidian

npx server

Obsidian, vault AI-Vault

on this PC

SFTP, email and alibaba-model-studio also keep a local copy in AI Output\.

Local AI, voice and vision

Everything that creates audio, images or models, plus the servers that read skills and screens.

lemonade

Python server

Lemonade at 192.168.12.174:13305

another machine on your network

LM-Studio-Files\

images\lemonade · audio\lemonade · models\lemonade

local-tts

Python server

Piper

piper.exe plus voices folder

LM-Studio-Files\audio\tts

speech files

alibaba-model-studio

free models only

Alibaba DashScope (intl)

OpenAI-compatible endpoint

AI Output\

results

application-intelligence

Python server

application-skills\

per-app skill files

computer-vision

Python server inferred

computer-vision\

computer_vision_mcp.py

Lemonade logs and memory

lemonade_learning_memory.json
what the server has learned across jobs
lemonade_tool_calls.jsonl / .md
every Lemonade call
mcp_tool_calls_all.jsonl
global tool-call log
mcp_task_journal.md
shared task journal, also used by the task engine

Folders and files

Everything under C:\Users\awise\Desktop\LMStudio-MCP-Transfer, with the servers that touch each part.

Folders
  • AI Output\alibaba, sftp, email and task-engine results
  • AI-Memory\
    • long-term-memory\dataChromaDB, embeddings, backups
    • task-engine\dataqueue and state
  • LM-Studio-Files\
    • audio\ttsPiper output, voices\ and piper\piper\piper.exe
    • audio\lemonade
    • images\lemonade
    • models\lemonade
    • task-engine\markdown_tasks
  • lmstudio-config\
    • mcp.json
    • server-logs\web_search_mcp.log, wordpress-mcp.log
  • Maton-MCP\server scripts, lemonade logs, journals
    • long-term-memory-mcp\shared .venv, long_term_memory_mcp.py
    • task_engine_data\
    • terminal_bridge_outputs\
    • voice_webpanel\
    • wiseai_stack_stats\
    • sftp_outputs\
  • web-search-mcp\src, dist, node_modules
  • wordpress-mcp-runtime\node_modules with mcp-wordpress-remote
  • application-skills\
  • computer-vision\
  • __pycache__\
Loose files in the root
  • application_intelligence_mcp.pyserver script, launched from here
  • ai_vault_mcp_check.mjscheck script for AI-Vault
  • wordpress-mcp-check.mjscheck script for WordPress
  • fix_json.py
  • cve_2026_99999_quick_check.py
  • cve_2026_99999_tester.py
  • all_states.txt
  • anything llm mcp file.txt
  • New Text Document.txt
  • PATH_MAPPING_REFERENCE.txt
  • README_SETUP_ON_NEW_PC.md
  • SYSTEM_GRAPH.md

Check scripts and notes are not started by LM Studio.

Maton-MCP scripts

From SYSTEM_GRAPH.md. Some of these scripts are not in mcp.json today: the voice scripts, the stack-stats server and both registry files.

Task engine stack

maton_service_registry_complete.py

plus maton_service_registry_ext.py

task_engine_supervisor.py

task_engine_worker.py

task_engine_mcp.py

reads task_engine_data\markdown_tasks

Voice

voice_server_control.py

voice_webpanel_server.py

output in voice_webpanel\ and LM-Studio-Files\audio\tts

Terminal and stats

terminal_bridge_mcp.py

reads mcp_lmstudio_*.json, writes terminal_bridge_outputs\

wiseai_stack_stats_server.py

reads mcp.json, writes wiseai_stack_stats\

Memory

long_term_memory_mcp.py

ChromaDB

plus memory_bridge_learning_memory.json

Move to a new PC

The transfer package holds five folders: Maton-MCP, web-search-mcp, AI-Memory, LM-Studio-Files and lmstudio-config.

Before you start

  • LM Studio installed
  • Python 3.10+ (same minor version, 64-bit Windows)
  • Node.js 18+
  • Git (optional)

Steps

  1. Copy the folders

    Put them under one base folder, for example C:\Maton-MCP-Transfer, keeping the five folder names.

  2. Fix paths in mcp.json

    Find and replace every old path using the table below. VS Code or Notepad++ handles it in one pass.

  3. Check the Python venv

    The copied .venv may work if Python version, 64-bit and Windows all match. If not, rebuild it:

    cd C:\Maton-MCP-Transfer\Maton-MCP\long-term-memory-mcp
    Remove-Item -Recurse -Force .venv
    python -m venv .venv
    .\.venv\Scripts\pip install -r requirements.txt
  4. Install Node parts

    In web-search-mcp run npm install then npm run build. The five npx servers install themselves on first run.

  5. Load the config in LM Studio

    Settings, then MCP. Import the edited mcp.json or add servers one by one.

  6. Test each server

    Start LM Studio and open the Logs tab. Fix any server that errors.

Path replacements

C:\Users\awise\Desktop\Maton MCP
C:\Maton-MCP-Transfer\Maton-MCP
C:\Users\awise\Downloads\web-search-mcp-main\web-search-mcp-main
C:\Maton-MCP-Transfer\web-search-mcp
C:\Users\awise\Documents\AI Memory
C:\Maton-MCP-Transfer\AI-Memory
C:\Users\awise\Documents\LM Studio Files
C:\Maton-MCP-Transfer\LM-Studio-Files
C:\Users\awise\.lmstudio\server-logs
C:\Maton-MCP-Transfer\lmstudio-config\server-logs
C:\Users\awise\Desktop, Documents, Downloads
C:\Users\<NEW_USERNAME>\Desktop, Documents, Downloads
C:\Users\awise
C:\Users\<NEW_USERNAME>
Note. The README's "old" paths do not match the current mcp.json, which already points at Desktop\LMStudio-MCP-Transfer. When you edit on the new PC, replace what is actually in the file, longest paths first.

Original machine

User
awise, Windows
Live LM Studio config
C:\Users\awise\.lmstudio\mcp.json
Package built
2026-10-22

Credentials and access

Secrets are stored as plain text in mcp.json. This page shows none of them.

Servers with stored secrets · 10
  • long_term_memory
  • maton-gateway
  • alibaba-model-studio
  • aaronwiseai-sftp
  • aaronwiseai-business-email
  • obsidian
  • wordpress
  • AI-Vault
  • context7
  • huggingface-explicit-only
Folder reach
filesystem
LMStudio-MCP-Transfer, Desktop, Documents, Downloads
terminal-bridge
Maton-MCP, LM-Studio-Files, Desktop, Documents, Downloads. CMD off.
Rotate before sharing. The README says the same: delete or rotate these keys after a transfer, especially if the USB drive travels. Environment variables or a secrets manager keep them out of the file.