Azure AI Foundry MCP Server (Experimental Interface)

This implementation serves as a Model Context Protocol server interfacing directly with Azure AI Foundry, offering a unified access layer for generative models, knowledge indexing utilities, performance assessment frameworks, and more.

Available Functionalities (Tool Sets)

Feature Set: Models

Domain	Utility Name	Purpose Description
Discovery	list_models_from_model_catalog	Fetches the roster of accessible models documented in the Azure AI Foundry asset registry.
	list_azure_ai_foundry_labs_projects	Acquires listings for cutting-edge AI artifacts curated by Microsoft Research within the Foundry Labs environment.
	get_model_details_and_code_samples	Retrieves comprehensive specification sheets and illustrative code snippets for a specific catalog entry.
Development	get_prototyping_instructions_for_github_and_labs	Supplies detailed guidance and prerequisite setup walkthroughs for commencing work with Foundry and Labs models.
Deployment	get_model_quotas	Queries the established usage limits for model instantiation within a designated Azure region.
	create_azure_ai_services_account	Provisions a new operational instance of Azure AI Services.
	list_deployments_from_azure_ai_services	Obtains a manifest of currently active deployments hosted on Azure AI Services.
	deploy_model_on_ai_services	Initiates the deployment sequence of a specified model onto the Azure AI Services infrastructure.
	create_foundry_project	Establishes a fresh organizational container (project) within Azure AI Foundry.

Feature Set: Knowledge Base Management

Domain	Utility Name	Purpose Description
Indexing	list_index_names	Returns all registered names associated with the backing AI Search Service indexes.
	list_index_schemas	Retrieves the structural definitions (schemas) for every index within the AI Search Service.
	retrieve_index_schema	Fetches the precise definition for one particular index from the Search Service.
	create_index	Constructs a novel index structure.
	modify_index	Applies alterations to the definition metadata of an existing index structure.
	delete_index	Permanently removes a specified index resource.
Data Item	add_document	Ingests a singular data item into the target index.
	delete_document	Erases a specific data item from the index store.
Searching	query_index	Executes a search operation against a specified index to yield relevant data records.
	get_document_count	Reports the total cardinality of items currently housed within the index.
Automation	list_indexers	Retrieves the names of all configured indexer processes linked to the AI Search Service.
	get_indexer	Fetches the complete configuration blueprint for a nominated indexer instance.
	create_indexer	Generates a new indexer job, linking a data source, target index, and processing skillset.
	delete_indexer	Removes an indexer definition from the Search Service registry.
Source Mgmt	list_data_sources	Lists the defined data source connections utilized by the AI Search Service.
	get_data_source	Retrieves the complete connection parameters for a specified data source definition.
Skill Sets	list_skill_sets	Returns the identifiers for all defined processing skill sets within the service.
	get_skill_set	Fetches the detailed operational composition of a chosen skill set.
Content Access	fk_fetch_local_file_contents	Reads and returns the data payload from a specified path on the local filesystem (e.g., configuration, text samples).
	fk_fetch_url_contents	Retrieves and returns the content accessible via a remote network Uniform Resource Locator (URL).

Feature Set: Evaluation Framework

Domain	Utility Name	Purpose Description
Utility Kit	list_text_evaluators	Enumerates all registered text quality assessment modules.
	list_agent_evaluators	Enumerates all registered conversational agent performance assessment modules.
	get_text_evaluator_requirements	Displays the required input parameters necessary for each text assessment utility.
	get_agent_evaluator_requirements	Displays the necessary input payload structure for each agent performance assessment utility.
Text Assessment	run_text_eval	Executes one or multiple text quality evaluators against content provided either as a JSONL file or direct input.
	format_evaluation_report	Transforms raw evaluation outcomes into a human-readable Markdown summary document.
Agent Assessment	agent_query_and_evaluate	Submits a query to a specified agent and subsequently evaluates the resulting output using selected metrics. (Complete lifecycle assessment).
	run_agent_eval	Assesses a singular agent invocation event using defined parameters (query, reply, tool usages, service definitions).
Agent Interface	list_agents	Lists every registered Azure AI Agent instance within the active configuration context.
	connect_agent	Sends an operational query directly to a designated agent endpoint.
	query_default_agent	Submits a query to the agent designated as the system default via environment settings.

Feature Set: Model Customization (Finetuning)

Domain	Utility Name	Purpose Description
Tuning Jobs	fetch_finetuning_status	Retrieves exhaustive status reports and metadata for a fine-tuning execution, encompassing state, model utilized, timestamps, hyperparameter settings, and error logs.
	list_finetuning_jobs	Generates a list of all ongoing or completed tuning operations within the resource boundary, showing IDs and current states for oversight.
	get_finetuning_job_events	Fetches a time-ordered sequence of all significant occurrences pertinent to a specific tuning task (training steps, validation checks, finalization).
	get_finetuning_metrics	Extracts performance metrics, such as loss curves and accuracy scores, for a specified tuning run to support analysis.
	list_finetuning_files	Lists all data assets currently staged for use in Azure OpenAI fine-tuning operations, including identifiers, names, roles, and readiness status.
	execute_dynamic_swagger_action	Invokes any operational endpoint dynamically exposed via a provided Swagger/OpenAPI specification for flexible service interaction.
	list_dynamic_swagger_tools	Discovers and lists all operational utilities registered dynamically based on an OpenAPI specification, enabling automation discovery.

Illustrative Interaction Scenarios

Models

Model Exploration

• How may this system assist me in identifying a suitable model?
• Which model assets are accessible through the Azure AI Foundry infrastructure?
• List the OpenAI models currently exposed via Azure AI Foundry.
• Identify the ten most frequently utilized models within Azure AI Foundry.
• Which models exhibit strong logical reasoning capabilities? Please segment the results into two groups: large-scale and compact models.
• Provide a comparative analysis detailing the distinctions between the Phi model variants.
• Display the official model card documentation for Phi-4-reasoning.
• Can you illustrate the procedure for benchmarking a model's performance?
• What designation is given to the cost-free testing environment within Azure AI Foundry?
• Is it permissible to employ a personal GitHub authentication token for model evaluation?
• Detail the most recent model versions compatible with GitHub token authentication.
• Who are the entities responsible for publishing the models available in Azure AI Foundry?
• Show me the models released by the Meta organization.
• List models whose usage is governed by the MIT license agreement.

Prototype Construction

• Explain how this system facilitates the creation of a preliminary application (prototype) utilizing a model.
• Detail the process for constructing a prototype leveraging an OpenAI model authenticated via my GitHub credentials. Do not proceed with actual creation yet.
• Suggest several practical use cases for building prototypes around these models.
• Offer background information regarding the Azure AI Foundry Labs environment.
• Provide context on the 'Omniparser' utility and its potential applications.
• Can you guide me through building a prototype that incorporates the Omniparser functionality?

OpenAI Model Deployment

• Outline the steps required to deploy an OpenAI model via this interface.
• What sequence of actions is necessary to instantiate OpenAI models using Azure AI Foundry resources?
• Clarify the method for utilizing OpenAI models in Azure AI Foundry via a GitHub token. Is this suitable for high-stakes production environments?
• If I possess an active Azure AI Services resource, can I use it for deploying OpenAI models?
• Define the concept of 'quota' as it applies to OpenAI models within Azure AI Foundry.
• Retrieve the current usage limitations applicable to my assigned AI Services resource.

Getting Started: Rapid Integration with GitHub Copilot

Utilizing This GitHub Template, which contains the minimal necessary MCP server configuration and dependencies, enables swift bootstrapping for personalized projects.

This integration automatically configures the MCP server within your VS Code user settings. Note that the uvx utility must be present in your operating environment to execute the server process.

On-Premises Setup Instructions

1. Install the uv package manager by referencing Installation Guide for uv.
2. Start a new development workspace session within VS Code.
3. (Optional) To define persistent environment variables, establish a .env file in the top-level directory of your workspace.

4. Construct the configuration file at .vscode/mcp.json within the workspace root.

   { "servers": { "mcp_foundry_server": { "type": "stdio", "command": "uvx", "args": [ "--prerelease=allow", "--from", "git+https://github.com/azure-ai-foundry/mcp-foundry.git", "run-azure-ai-foundry-mcp", "--envFile", "${workspaceFolder}/.env" ] } } }

5. Activate the server instance by clicking the 'Start' control associated with the server definition in .vscode/mcp.json.

6. Switch GitHub Copilot to Agent mode and commence interaction.

Refer to Advanced Client Setup Documentation for more granular detail on configuring the MCP server.

Secure Configuration via Environment Variables

Environment variables are the prescribed mechanism for securely injecting sensitive parameters such as access credentials, service endpoints, and other private configurations into the MCP server runtime. This is crucial for any utility requiring external service authentication.

These variables should be defined within a .env file situated at the project's root. You reference this file during the MCP Server setup step, ensuring the server loads these variables upon initialization.

Consult the Environment Variable Example for a template structure.

Context	Parameter Name	Mandate?	Purpose Summary
Model Ops	GITHUB_TOKEN	Optional	Authenticates against the GitHub API for rate-limited model testing.
Knowledge	AZURE_AI_SEARCH_ENDPOINT	Mandatory	The fully qualified URI for the Azure AI Search instance (e.g., https://<name>.search.windows.net/).
	AZURE_AI_SEARCH_API_VERSION	Optional	Specifies the Search API version; defaults to 2025-03-01-preview.
	SEARCH_AUTHENTICATION_METHOD	Mandatory	Selection between service-principal or api-search-key.
	AZURE_TENANT_ID	Required for Service Principal Auth	The unique identifier for your Azure Active Directory tenant.
	AZURE_CLIENT_ID	Required for Service Principal Auth	The application registration identifier (Service Principal ID).
	AZURE_CLIENT_SECRET	Required for Service Principal Auth	The secret credential associated with the Service Principal identity.
	AZURE_AI_SEARCH_API_KEY	Required for Key Auth	The access key credential for the Azure AI Search service.
Assessment	EVAL_DATA_DIR	Mandatory	Directory path pointing to the JSONL input data file for evaluations.
	AZURE_OPENAI_ENDPOINT	Needed for Text Quality Assessors	The base URL endpoint for the Azure OpenAI service.
	AZURE_OPENAI_API_KEY	Needed for Text Quality Assessors	The secret key credential for accessing Azure OpenAI.
	AZURE_OPENAI_DEPLOYMENT	Needed for Text Quality Assessors	The specific model deployment name (e.g., gpt-4o).
	AZURE_OPENAI_API_VERSION	Needed for Text Quality Assessors	The version string for the OpenAI API invocation.
	AZURE_AI_PROJECT_ENDPOINT	Required for Agent Services	The endpoint URL utilized for interacting with Azure AI Agent services.

[!IMPORTANT] Summary Notes Model Operations - GITHUB_TOKEN facilitates GitHub API interactions for testing models; it is not obligatory when only browsing the Foundry catalog.

Knowledge Operations - Provisioning a Search Service is detailed in Azure Search Service Creation. - Azure AI Search supports heterogeneous credential methods. You may employ either Microsoft Entra ID authentication or Key-based authentication based on organizational security mandates and deployment context. - Review Search Authentication Overview for comprehensive details on supported credential mechanisms.

Evaluation Operations - For utilizing agent-based utilities or safety validators, ensure the associated Azure project credentials are fully valid and accessible. - If the scope is limited strictly to text quality assessment, providing the OpenAI endpoint and key credentials suffices.

Software Licensing

This project is distributed under the MIT License. License terms are fully detailed in the LICENSE file.

WIKIPEDIA: Tools for business management encompass all the systems, software applications, control mechanisms, computational methodologies, and operational frameworks employed by enterprises to effectively navigate shifting market dynamics, sustain competitive advantage, and elevate organizational performance metrics.

== Conceptual Framework == Management utilities can be segmented according to organizational function. This classification includes, but is not limited to, tools for forecasting, process orchestration, record-keeping, human capital management, strategic assessment, and performance monitoring. Key functional categories generally cover:

Utilities handling data entry and integrity verification across departments. Frameworks designed for monitoring and refining operational pipelines. Systems dedicated to aggregating information for high-level decision support. The rapid evolution of management technology over the past decade, driven by fast-paced technological advancement, makes selecting optimal business tools for specific organizational needs increasingly complex. This difficulty stems from perpetual pressures to reduce operational expenditure, maximize revenue generation, deeply understand client requirements, and deliver requisite products precisely as demanded. In this context, executives must adopt a strategic posture regarding business management software selection, prioritizing tailored integration over indiscriminate adoption of the newest releases. Misapplied tools often result in organizational instability; thus, business management systems demand careful selection followed by thorough customization to organizational protocols.

== Prominent Tools (2013 Benchmark) == Research conducted by Bain & Company in 2013 mapped the global utilization of business tools, reflecting regional market conditions and corporate requirements. The top ten instruments identified were:

Strategic planning schema Client relationship management platforms Employee satisfaction measurement instruments Performance benchmarking techniques Balanced scorecard methodologies Core competency identification frameworks Outsourcing strategies Organizational change management initiatives Supply chain oversight systems Mission and vision definition processes Market segmentation analysis Total Quality Management (TQM) protocols

== Business Software Applications == Business software, defined as a set of programs utilized by personnel to execute diverse corporate operations, serves to enhance productivity metrics, quantify performance outputs, and ensure precision in multifaceted business tasks. The evolution progressed from early Management Information Systems (MIS) to comprehensive Enterprise Resource Planning (ERP) suites, later incorporating Customer Relationship Management (CRM), culminating in the current domain of cloud-based enterprise management solutions. Value addition beyond IT investment relies critically on two factors: the efficacy of the deployment process and the judicious selection and customization of the utilized tools.

== Focus on Small and Medium Enterprises (SMEs) == Tools tailored for SMEs are vital as they provide mechanisms to conserve resources and...