feat: add --model CLI parameter for model selection by chindris-mihai-alexandru · Pull Request #19 · Cerebras/cerebras-code-mcp

chindris-mihai-alexandru · 2026-01-03T22:14:07Z

Summary

Adds the ability to select which Cerebras model to use via a --model CLI parameter, addressing the feature request in #6.

Changes

New Features

--model <model-name> / -m <model-name> - Select which Cerebras model to use
--help / -h - Display usage information and list of available models
Model validation with fallback to default if invalid model specified
Clear indication of which model is being used on server startup (regardless of selection method)

Implementation Details

Added AVAILABLE_MODELS constant in src/config/constants.js with all 6 supported Cerebras models
Added setModel() function to enable runtime model configuration
Used Node.js built-in parseArgs (node:util) for CLI argument parsing
Maintained backward compatibility - existing behavior unchanged if no --model flag provided
Updated Node.js requirement to >=18.3.0 for parseArgs support

Documentation

Updated README.md with:
- Model selection section
- List of all available models with parameters and speed
- Usage examples
- Reference to --help flag

Available Models

Model ID	Name	Parameters	Speed
`llama3.1-8b`	Llama 3.1 8B	8 billion	~2200 tokens/s
`llama-3.3-70b`	Llama 3.3 70B	70 billion	~2100 tokens/s
`gpt-oss-120b`	OpenAI GPT OSS	120 billion	~3000 tokens/s
`qwen-3-32b`	Qwen 3 32B	32 billion	~2600 tokens/s
`qwen-3-235b-a22b-instruct-2507`	Qwen 3 235B Instruct	235 billion	~1400 tokens/s (Preview)
`zai-glm-4.6`	Z.ai GLM 4.6	357 billion	~1000 tokens/s (Preview, Default)

Usage Examples

# Use Llama 3.3 70B
cerebras-mcp --model llama-3.3-70b

# Use Qwen 3 32B (shorthand)
cerebras-mcp -m qwen-3-32b

# Display help
cerebras-mcp --help

Testing

Verified --help displays correct information
Tested model selection with valid model names
Confirmed invalid model names fall back to default with warning
Verified backward compatibility (no flags = default behavior)
Server starts successfully with selected model
Model display shows on startup regardless of selection method

Copilot AI Review

All 5 automated review comments have been addressed in commit b6a21fc:

Added startup log showing active model
Hardcoded default model in help text for clarity
Fixed parameter formatting consistency in README
Updated Node.js requirement to >=18.3.0
Corrected GitHub URL to match package.json

Resolves

Closes #6

Notes

The CEREBRAS_MODEL environment variable still works and takes precedence if set
Default model remains zai-glm-4.6 to maintain existing behavior
No breaking changes - fully backward compatible

Implements feature requested in issue Cerebras#6 Changes: - Add --model/-m CLI parameter to specify Cerebras model at runtime - Add --help/-h flag to display usage information and available models - Add AVAILABLE_MODELS constant with all supported Cerebras models - Add setModel() function to enable runtime model configuration - Update README with model selection documentation Available models: - llama3.1-8b (8B, ~2200 tokens/s) - llama-3.3-70b (70B, ~2100 tokens/s) - gpt-oss-120b (120B, ~3000 tokens/s) - qwen-3-32b (32B, ~2600 tokens/s) - qwen-3-235b-a22b-instruct-2507 (235B, ~1400 tokens/s, Preview) - zai-glm-4.6 (357B, ~1000 tokens/s, Preview, Default) Usage: cerebras-mcp --model llama-3.3-70b cerebras-mcp -m qwen-3-32b cerebras-mcp --help Resolves Cerebras#6

Copilot

Pull request overview

This PR adds CLI model selection functionality to the Cerebras Code MCP Server, allowing users to specify which Cerebras model to use via a --model flag. The implementation uses Node.js's built-in parseArgs for argument handling and includes comprehensive help documentation.

Key Changes:

Added --model/-m CLI parameter for runtime model selection with validation and fallback to default
Implemented --help/-h flag displaying usage information and available models
Created AVAILABLE_MODELS constant with metadata for all 6 supported Cerebras models

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
src/index.js	Implements CLI argument parsing with parseArgs, model validation logic, and comprehensive help text
src/config/constants.js	Adds AVAILABLE_MODELS constant with model metadata and setModel() function for runtime configuration
README.md	Documents model selection feature, lists available models with specifications, and provides usage examples

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-03T22:17:46Z

    console.error('Cerebras Code MCP Server starting...');
    console.error(`📝 Debug logs will be written to: ${LOG_FILE}`);


The server startup should display which model is currently being used, regardless of whether it was set via CLI flag or environment variable. Currently, the model is only shown when explicitly set via the --model flag. Add a log statement after line 95 to always show the active model, similar to how API key availability is logged.

Copilot · 2026-01-03T22:17:46Z

+Options:
+  -c, --config              Run interactive configuration wizard
+  -r, --remove              Remove MCP server configuration
+  -m, --model <model-name>  Specify which Cerebras model to use (default: ${config.cerebrasModel})


The help message displays the default model by accessing config.cerebrasModel at template string evaluation time (line 54). However, if the CEREBRAS_MODEL environment variable is set, this will show the environment variable's value as the default, which could be confusing. Consider hardcoding the default as "zai-glm-4.6" in the help text, and separately mention that it can be overridden by the CEREBRAS_MODEL environment variable.

Suggested change

-m, --model <model-name> Specify which Cerebras model to use (default: ${config.cerebrasModel})

-m, --model <model-name> Specify which Cerebras model to use (default: zai-glm-4.6)

This default can be overridden via the CEREBRAS_MODEL environment variable.

Copilot · 2026-01-03T22:17:47Z

+- `llama-3.3-70b` - Llama 3.3 70B (70 billion parameters, ~2100 tokens/s)
+- `gpt-oss-120b` - OpenAI GPT OSS (120 billion parameters, ~3000 tokens/s)
+- `qwen-3-32b` - Qwen 3 32B (32 billion parameters, ~2600 tokens/s)
+- `qwen-3-235b-a22b-instruct-2507` - Qwen 3 235B Instruct (235 billion, ~1400 tokens/s, Preview)


Inconsistent formatting in the model descriptions. Line 51 says "235 billion" while other entries say "235 billion parameters". Line 52 also says "357 billion parameters" while line 51 drops the "parameters" word. For consistency, either all entries should include "parameters" or all should omit it.

Suggested change

- `qwen-3-235b-a22b-instruct-2507` - Qwen 3 235B Instruct (235 billion, ~1400 tokens/s, Preview)

- `qwen-3-235b-a22b-instruct-2507` - Qwen 3 235B Instruct (235 billion parameters, ~1400 tokens/s, Preview)

Copilot · 2026-01-03T22:17:47Z

 */

-import { config, debugLog, LOG_FILE } from './config/constants.js';
+import { parseArgs } from 'node:util';


The parseArgs function from node:util was stabilized in Node.js 18.3.0, but package.json specifies "node": ">=18.0.0". Users running Node.js 18.0.0-18.2.x will encounter errors because parseArgs is experimental in those versions and requires a flag to use. Consider either updating the minimum Node.js version to ">=18.3.0" in package.json, or using the util.parseArgs with the experimental flag check for older versions.

Copilot · 2026-01-03T22:17:47Z

+  cerebras-mcp -m qwen-3-32b                   # Use Qwen 3 32B (shorthand)
+  cerebras-mcp --config                        # Configure API keys
+
+For more information, visit: https://github.com/Cerebras/cerebras-code-mcp


The GitHub repository URL in the help text does not match the actual repository URL specified in package.json. The help text references "https://github.com/Cerebras/cerebras-code-mcp" but package.json indicates the repository is at "https://github.com/kevint-cerebras/cerebras-mcp". Update the URL to match the correct repository location.

Suggested change

For more information, visit: https://github.com/Cerebras/cerebras-code-mcp

For more information, visit: https://github.com/kevint-cerebras/cerebras-mcp

- Add startup log showing active model regardless of selection method - Hardcode default model (zai-glm-4.6) in help text for clarity - Fix parameter formatting: '235 billion' → '235 billion parameters' - Update Node.js requirement to >=18.3.0 for parseArgs support - Fix GitHub URL to match package.json repository

chindris-mihai-alexandru · 2026-01-03T22:28:32Z

Copilot AI Review Feedback Addressed

I've addressed all 5 comments in commit b6a21fc:

Changes Made:

Model Display Enhancement - Added startup log showing the active model regardless of how it was set (CLI flag, env var, or default)
- Now displays: Using model: Z.ai GLM 4.6 (zai-glm-4.6)
Help Text Clarity - Hardcoded zai-glm-4.6 as the default model in help text instead of using the dynamic variable
- Makes the help output clearer and doesn't depend on runtime state
Parameter Formatting Consistency - Fixed 235 billion → 235 billion parameters in README.md
- Now consistent with other model descriptions
Node.js Version Requirement - Updated package.json engines to require Node.js >=18.3.0
- This is when parseArgs was stabilized (no longer experimental)
GitHub URL Correction - Fixed help text URL to match package.json repository
- Changed from github.com/Cerebras/cerebras-code-mcp to github.com/kevint-cerebras/cerebras-mcp

All changes maintain backward compatibility and improve code quality.

chindris-mihai-alexandru · 2026-01-03T22:41:13Z

Model List Verification

I've verified that the model list in this PR matches the current official Cerebras Inference API documentation (January 2026).

Models Included (6 total)

Production Models:

llama3.1-8b - Llama 3.1 8B (8 billion parameters, ~2200 tokens/s)
llama-3.3-70b - Llama 3.3 70B (70 billion parameters, ~2100 tokens/s)
gpt-oss-120b - OpenAI GPT OSS (120 billion parameters, ~3000 tokens/s)
qwen-3-32b - Qwen 3 32B (32 billion parameters, ~2600 tokens/s)

Preview Models:

qwen-3-235b-a22b-instruct-2507 - Qwen 3 235B Instruct (235 billion parameters, ~1400 tokens/s)
zai-glm-4.6 - Z.ai GLM 4.6 (357 billion parameters, ~1000 tokens/s, Default)

Source

Official documentation: https://inference-docs.cerebras.ai/models

Note on qwen-3-coder-480b

The qwen-3-coder-480b model mentioned in issue #6 was officially deprecated by Cerebras on November 5, 2025 (source: Deprecations). Cerebras recommends migrating to zai-glm-4.6 as a replacement. This PR uses only currently supported models to ensure long-term reliability.

chindris-mihai-alexandru · 2026-02-15T18:07:41Z

Closing this stale PR to keep my open queue focused. If maintainers still want this change, I can reopen and refresh it.

Copilot AI review requested due to automatic review settings January 3, 2026 22:14

chindris-mihai-alexandru requested a review from kevint-cerebras as a code owner January 3, 2026 22:14

Copilot started reviewing on behalf of chindris-mihai-alexandru January 3, 2026 22:14 View session

Copilot AI reviewed Jan 3, 2026

View reviewed changes

chindris-mihai-alexandru mentioned this pull request Jan 3, 2026

Security: Namespace MCP tool name to prevent collision (write → cerebras_write) #20

Closed

chindris-mihai-alexandru closed this Feb 15, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: add --model CLI parameter for model selection#19

feat: add --model CLI parameter for model selection#19
chindris-mihai-alexandru wants to merge 2 commits intoCerebras:mainfrom
chindris-mihai-alexandru:feature/model-selection-cli

chindris-mihai-alexandru commented Jan 3, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jan 3, 2026

Uh oh!

Copilot AI Jan 3, 2026

Uh oh!

Copilot AI Jan 3, 2026

Uh oh!

Copilot AI Jan 3, 2026

Uh oh!

Copilot AI Jan 3, 2026

Uh oh!

chindris-mihai-alexandru commented Jan 3, 2026 •

edited

Loading

Uh oh!

chindris-mihai-alexandru commented Jan 3, 2026 •

edited

Loading

Uh oh!

chindris-mihai-alexandru commented Feb 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		console.error('Cerebras Code MCP Server starting...');
		console.error(`📝 Debug logs will be written to: ${LOG_FILE}`);

	-m, --model <model-name> Specify which Cerebras model to use (default: ${config.cerebrasModel})
	-m, --model <model-name> Specify which Cerebras model to use (default: zai-glm-4.6)
	This default can be overridden via the CEREBRAS_MODEL environment variable.

	- `qwen-3-235b-a22b-instruct-2507` - Qwen 3 235B Instruct (235 billion, ~1400 tokens/s, Preview)
	- `qwen-3-235b-a22b-instruct-2507` - Qwen 3 235B Instruct (235 billion parameters, ~1400 tokens/s, Preview)

	For more information, visit: https://github.com/Cerebras/cerebras-code-mcp
	For more information, visit: https://github.com/kevint-cerebras/cerebras-mcp

Conversation

chindris-mihai-alexandru commented Jan 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

New Features

Implementation Details

Documentation

Available Models

Usage Examples

Testing

Copilot AI Review

Resolves

Notes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 3, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 3, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 3, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 3, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 3, 2026

Choose a reason for hiding this comment

Uh oh!

chindris-mihai-alexandru commented Jan 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Copilot AI Review Feedback Addressed

Changes Made:

Uh oh!

chindris-mihai-alexandru commented Jan 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Model List Verification

Models Included (6 total)

Source

Note on qwen-3-coder-480b

Uh oh!

chindris-mihai-alexandru commented Feb 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chindris-mihai-alexandru commented Jan 3, 2026 •

edited

Loading

chindris-mihai-alexandru commented Jan 3, 2026 •

edited

Loading

chindris-mihai-alexandru commented Jan 3, 2026 •

edited

Loading