Xplainable

You run the Xplainable MCP Server to securely expose Xplainable AI platform capabilities through a standardized MCP interface. It provides authenticated read and write operations, input validation, rate limiting, and audit logging so you can manage models, deployments, preprocessors, and reports from your MCP client with confidence.

How to use

To connect your MCP client to the Xplainable MCP Server, choose a runtime method that suits your setup. You can run the server locally for development or connect a remote MCP client using one of the provided configurations. The server exposes read operations for models, deployments, preprocessors, and collections, and write operations for deploying models, managing deployments, and generating reports when you have the proper authorization.

How to install

Prerequisites you need before installation include Python and pip on your system. You will run a Python-based MCP server that you can start with a simple command.

Step 1: Install the MCP server package using Python’s package manager.

pip install xplainable-mcp-server

Additional configuration and usage notes

Environment variables you may set for backend authentication include the API key and host for the Xplainable platform. You can configure these when running the server or in an environment file.

Starting the server for development (localhost only) or production (with TLS/proxy) is done via the MCP launcher.

# For development (localhost only)
xplainable-mcp

# For production (with TLS/proxy)
xplainable-mcp --host 0.0.0.0 --port 8000

Security and operational notes

Authentication uses token-based access for MCP clients and API keys for the Xplainable backend. The server enforces transport security, rate limiting, and audit logging to track operations.

Troubleshooting

If you encounter connection issues, verify that your API key and host are correctly set in the environment. Check that the server process is reachable from your MCP client and confirm the client’s configuration matches the server’s supported tools and endpoints.

Examples and workflows

You can perform common workflows such as listing models, deploying a model version, generating a deployment key, and testing inference through your MCP client by following the relevant write/read tools exposed by the server. Ensure you have the necessary permissions for write operations.

Available tools

list_tools

Discover all available MCP tools with descriptions and parameters.

get_connection_info

Retrieve connection and diagnostic information about the MCP server.

list_team_models

List all models for a given team or all teams if no ID is provided.

get_model

Get detailed information about a specific model by its ID.

list_model_versions

List all versions for a specific model.

list_deployments

List all deployments for teams or a specific scope.

list_preprocessors

List all preprocessors available in the environment.

get_preprocessor

Get details for a specific preprocessor.

get_collection_scenarios

List scenarios within a particular collection.

get_active_team_deploy_keys_count

Get the count of active deploy keys for a team.

misc_get_version_info

Retrieve version information of the MCP server and related tooling.

activate_deployment

Activate a deployment (requires write access).

deactivate_deployment

Deactivate a deployment (requires write access).

generate_deploy_key

Generate a deployment key for a deployment, with optional expiry description.

get_deployment_payload

Get a sample payload for a deployment.

gpt_generate_report

Generate a GPT-based report for a model/version.