An MCP server for MLflow that exposes tools to manage and inspect experiments, runs, registered models, MLflow 3 logged models, traces, assessments, webhooks, and prompt-optimization jobs. It also supports aggregation tools, MCP prompts/resources, and Databricks-only trace attachment access.
From the registry: MLflow MCP — experiments, runs, registered models, versions, traces, assessments (MLflow 3)
Please install the `mlflow` MCP server into my current AI client (that's you).
Required prerequisites (do these first if not already done):
- **Install Node.js 22+** — Node.js 22 or newer is required to run the server via npx.
- **Have access to an MLflow tracking server** — You need an MLflow tracking URI for an OSS MLflow server or a Databricks-managed MLflow workspace.
Optional prerequisites:
- Obtain MLflow authentication credentials if needed — If your MLflow server requires authentication, prepare either a bearer token or basic-auth credentials. For Databricks, use a PAT or service-principal token.
Canonical MCP server config (stdio transport):
- command: `npx`
- args: ["-y","@us-all/mlflow-mcp"]
- required environment variables:
- `MLFLOW_TRACKING_URI`: MLflow tracking URL for your MLflow server or Databricks workspace. (example: `<your-mlflow-tracking-uri>`)
- optional environment variables:
- `MLFLOW_TRACKING_TOKEN`: Bearer token for MLflow authentication. Use for Databricks PAT or other bearer-auth setups. (example: `<your-mlflow-tracking-token>`)
- `MLFLOW_TRACKING_USERNAME`: Basic-auth username for MLflow. Alternative to bearer token. (example: `<your-mlflow-username>`)
- `MLFLOW_TRACKING_PASSWORD`: Basic-auth password for MLflow. Used with MLFLOW_TRACKING_USERNAME. (example: `<your-mlflow-password>`)
- `MLFLOW_EXPERIMENT_ID`: Default experiment ID for tools that can use an implicit experiment. (example: `<your-experiment-id>`)
- `MLFLOW_ALLOW_WRITE`: Set to true to enable create, update, and delete operations. Defaults to false. (example: `<true-or-false>`)
- `MLFLOW_TOOLS`: Comma-separated allowlist of tool categories to enable. (example: `<experiments,runs,registry,traces>`)
- `MLFLOW_DISABLE`: Comma-separated denylist of tool categories to disable. Ignored if MLFLOW_TOOLS is set. (example: `<categories-to-disable>`)
- `MCP_TRANSPORT`: Transport mode. Defaults to stdio; set to http to enable Streamable HTTP transport instead. (example: `<stdio-or-http>`)
- `MCP_HTTP_TOKEN`: Bearer token required when MCP_TRANSPORT=http, unless auth is explicitly skipped. (example: `<your-mcp-http-token>`)
- `MCP_HTTP_PORT`: HTTP listen port when using HTTP transport. Defaults to 3000. (example: `<your-http-port>`)
- `MCP_HTTP_HOST`: HTTP bind host when using HTTP transport. Defaults to 127.0.0.1. (example: `<your-http-host>`)
- `MCP_HTTP_SKIP_AUTH`: Set to true to skip bearer auth for HTTP transport, such as when a reverse proxy handles authentication. (example: `<true-or-false>`)
Note: Default and preferred configuration uses stdio via npx. The server also supports Streamable HTTP when MCP_TRANSPORT=http, with POST /mcp and GET /health. If using HTTP mode, MCP_HTTP_TOKEN is required unless MCP_HTTP_SKIP_AUTH=true.
Add this MCP server to my current client's config in the correct format for you. If you need secrets or credentials I haven't provided, ASK me — do not invent values or leave raw placeholders. After adding it, tell me how to verify the server is connected.MLFLOW_TRACKING_URIrequiredMLflow tracking server URL or Databricks workspace URLMCP_HTTP_TOKENrequiredBearer token required when using HTTP transportAI orchestration with hive-mind swarms, neural networks, and 87 MCP tools for enterprise dev.
Persistent codebase knowledge graph. Survives session restarts and context compaction.
A powerful toolkit for coding, providing semantic retrieval and editing capabilities.