# us-all/mlflow-mcp-server [Health: Active]

**Category:** 🛠️ Other Tools and Integrations  
**Repository:** https://github.com/us-all/mlflow-mcp-server  
**GitHub Stars:** 1  
**npm Downloads (last month):** 232  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/us-all-mlflow-mcp-server

## Description
MLflow — 82 tools across experiments, runs, registered models, model versions, traces, assessments. MLflow 3 GenAI traces support and 5 workflow Prompts (analyze-failed-traces, promote-best-run, compare-top-runs).

## Tools
Capabilities this server exposes over MCP:

- **create-experiment**
- **search-experiments**
- **get-experiment**
- **get-experiment-by-name**
- **update-experiment**
- **delete-experiment**
- **restore-experiment**
- **set-experiment-tag**
- **delete-experiment-tag**
- **create-registered-model**
- **get-registered-model**
- **search-registered-models**
- **rename-registered-model**
- **update-registered-model**
- **delete-registered-model**
- **get-latest-model-versions**
- **set-registered-model-tag**
- **delete-registered-model-tag**
- **set-registered-model-alias**
- **delete-registered-model-alias**
- **get-model-version-by-alias**
- **create-model-version**
- **get-model-version**
- **search-model-versions**
- **update-model-version**
- **delete-model-version**
- **transition-model-version-stage**
- **get-model-version-download-uri**
- **set-model-version-tag**
- **delete-model-version-tag**
- **create-logged-model**
- **search-logged-models**
- **get-logged-model**
- **finalize-logged-model**
- **delete-logged-model**
- **set-logged-model-tags**
- **delete-logged-model-tag**
- **log-logged-model-params**
- **search-traces**
- **get-trace**
- **get-trace-info**
- **delete-traces**
- **set-trace-tag**
- **delete-trace-tag**
- **list-trace-attachments**
- **get-trace-attachment**
- **log-feedback**
- **log-expectation**
- **get-assessment**
- **update-assessment**
- **delete-assessment**
- **create-webhook**
- **list-webhooks**
- **get-webhook**
- **update-webhook**
- **delete-webhook**
- **test-webhook**
- **create-prompt-optimization-job**
- **get-prompt-optimization-job**
- **search-prompt-optimization-jobs**
- **cancel-prompt-optimization-job**
- **delete-prompt-optimization-job**

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "mlflow-mcp-server": {
    "command": "npx",
    "args": ["-y","@us-all/mlflow-mcp"],
    "env": {
      "MLFLOW_TRACKING_URI": "",
      "MLFLOW_TRACKING_TOKEN": "",
      "MLFLOW_TRACKING_USERNAME": "",
      "MLFLOW_TRACKING_PASSWORD": "",
      "MLFLOW_EXPERIMENT_ID": "",
      "MLFLOW_ALLOW_WRITE": "",
      "MLFLOW_TOOLS": "",
      "MLFLOW_DISABLE": ""
    }
  }
}
```

**Requires environment variables:** `MLFLOW_TRACKING_URI`, `MLFLOW_TRACKING_TOKEN`, `MLFLOW_TRACKING_USERNAME`, `MLFLOW_TRACKING_PASSWORD`, `MLFLOW_EXPERIMENT_ID`, `MLFLOW_ALLOW_WRITE`, `MLFLOW_TOOLS`, `MLFLOW_DISABLE` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What us-all/mlflow-mcp-server MCP server does

The us-all/mlflow-mcp-server MCP server exposes MLflow data and operations as MCP tools. Its coverage includes experiments, runs, model registry entities, MLflow 3 LoggedModel records, traces, assessments, webhooks, and prompt-optimization jobs. The package also provides four workflow prompts: `debug-failed-traces`, `promote-best-run`, `compare-top-runs`, and `annotate-trace-quality`.

The server is suited to agent-driven experiment analysis and MLflow 3 trace debugging. Example tasks include finding the best run by a metric, comparing hyperparameters across top runs, inspecting failed traces, reviewing feedback annotations, and following a registered model back to its source run.

## How it works

The server uses `MLFLOW_TRACKING_URI` to connect to an OSS MLflow deployment or a Databricks workspace. Bearer-token and basic authentication options are available through environment variables. It starts with stdio transport; setting `MCP_TRANSPORT=http` enables Streamable HTTP with `/mcp` for JSON-RPC requests and `/health` for liveness checks.

`summary-experiment` combines experiment details, top runs, and metric statistics in one response. `summarize-run` reduces repeated metric-history fields in large responses. Several read tools support `extractFields`, allowing a caller to request only selected response fields. The server also exposes six `mlflow://` resource URI patterns for runs, experiments, artifacts, registered model versions, and traces.

## Setup and configuration

Install it through the npm package with `npx -y @us-all/mlflow-mcp`, run the published Docker image, or build from source with pnpm. Claude Desktop and Claude Code examples are provided in the project documentation.

`MLFLOW_TRACKING_URI` is required. Use `MLFLOW_TRACKING_TOKEN` for a bearer token such as a Databricks personal access token, or provide `MLFLOW_TRACKING_USERNAME` and `MLFLOW_TRACKING_PASSWORD` for basic authentication. `MLFLOW_EXPERIMENT_ID` can supply a default experiment context.

Writes are disabled by default. Set `MLFLOW_ALLOW_WRITE=true` before using create, update, delete, restore, logging, registration, trace, assessment, webhook, or prompt-optimization mutations. For HTTP mode, configure `MCP_HTTP_TOKEN` unless authentication is intentionally skipped behind a trusted reverse proxy.

## Tools and capabilities

The us-all/mlflow-mcp-server MCP server groups tools into eight configurable categories: experiments, runs, registry, logged models, traces, assessments, webhooks, and prompts. `MLFLOW_TOOLS` allows an allowlist of categories, while `MLFLOW_DISABLE` provides a denylist when no allowlist is set.

Trace tools can search and retrieve MLflow 3 traces, and Databricks deployments additionally support listing and retrieving trace attachments. Run and experiment tools support metric, parameter, tag, artifact, and history operations. Registry tools cover registered models and model versions. A `search-tools` meta-tool helps locate available tools, and `compare-runs` can render a comparison card for supported ChatGPT clients while returning JSON content to Claude clients.

## Limitations and notes

Trace search limits each page to 500 results. Larger searches require following `nextPageToken`; the cap does not limit the total paginated result set. Trace attachments are Databricks-only because OSS MLflow returns 404 for those routes.

The documentation notes that bearer and basic authentication paths have not yet been validated against a live Databricks deployment. OSS MLflow 3.12 is supported with bearer authentication optional. HTTP mode requires a bearer token by default, and the server binds to `127.0.0.1` on port 3000 unless configured otherwise.

_Full upstream README: https://allmcps.com/mcp/us-all-mlflow-mcp-server/readme_