LM StudioMCPLocal ModelsIntermediate7 min read

Why this setup matters

Run private or local AI workflows without sending full conversation history to hosted assistants.
Keep Saudi market data structured and reliable through SAHMK MCP tools.
Useful for internal tooling, research workflows, and developer-side analysis loops.

What you need

LM Studio installed and running
SAHMK MCP installed
SAHMK API key
A local model loaded in LM Studio

Tested model examples

Gemma 4 26B E2B
Qwen 3.6 35B A3B

You can use other local models as long as tool calling is stable in your runtime.

LM Studio MCP setup (simple flow)

Install SAHMK MCP:

bash

pip install -U sahmk-mcp

Set your API key in the environment used by LM Studio:

bash

export SAHMK_API_KEY="shmk_live_your_key_here"

Add a stdio MCP server entry in LM Studio (exact UI labels can vary by version):

json

{
  "name": "sahmk",
  "transport": "stdio",
  "command": "sahmk-mcp",
  "env": {
    "SAHMK_API_KEY": "shmk_live_your_key_here"
  }
}

Select your local model, open a new chat, and verify MCP tools are available.

Recommended system prompt

Use this as your default system prompt for Saudi market analysis in LM Studio:

text

You are a Saudi market intelligence assistant.

Focus:

* Saudi stocks
* company snapshots
* dividends
* historical performance
* sectors
* market context
* recent news

Style:

* concise
* factual
* practical
* no fluff
* no dramatic metaphors
* explain clearly but briefly
* use tools often
* do not guess reasons for price moves unless supported by tool output or web results
* if uncertain, say so

Tools:

* Use Sahmk MCP for Saudi market data.
* Use Brave Search for fresh public news or external context.
* Prefer Sahmk MCP for prices, dividends, company data, historicals, and sectors.
* Prefer Brave Search only for recent developments, announcements, or public web context.

Language:

* Reply in the user’s language.
* Arabic should feel natural, clear, and investor-friendly.

Output:

* answer directly
* keep it compact by default
* expand only when the user asks for deeper analysis

Context length guidance

Default small contexts like 4K are often too limiting for multi-turn market workflows.
Use 64K context as a practical baseline if model/runtime/hardware support it.
128K+ is better for longer research sessions, multi-step comparisons, and tool-heavy chats.

Actual usable context depends on model quality, runtime behavior, and available hardware.

Example prompts