Trustwise

Trustwise is an MCP server that connects LLM clients to the Trustwise evaluation platform for automated assessment of model safety, alignment, and performance. AI engineers, researchers, and developers integrate this server to programmatically evaluate model outputs directly within compatible development environments like Claude Desktop and Cursor. The server interfaces with the Trustwise API through a Docker container, exposing a suite of specialized evaluation tools that calculate metrics across multiple dimensions of generated text. Users can run evaluations for output faithfulness to retrieved context, answer and context relevancy, and summarization quality to detect hallucinations and context loss. The server also provides critical security checks for prompt injection vulnerabilities and personally identifiable information exposure in generated content. Additionally, it evaluates stylistic and conversational attributes such as clarity, formality, helpfulness, tone, toxicity, simplicity, and adherence to specific operational policies or system instructions. For operational oversight and efficiency tracking, the server calculates inference stability across repeated completions and provides quantitative estimates for model carbon footprint and financial inference costs.

Category: AI & LLM Tooling

Tags: alignment, evaluation, performance, safety, trustwise

Visit Trustwise

How to install and configure Trustwise

1. Verify that Docker is installed and running on your system. 2. Obtain an API key from the Trustwise platform. 3. For Claude Desktop, edit your configuration file and add the server definition under mcpServers: json { "mcpServers": { "trustwise": { "command": "docker", "args": [ "run", "-i", "--rm", "-e", "TW_API_KEY", "ghcr.io/trustwiseai/trustwise-mcp-server:latest" ], "env": { "TW_API_KEY": "<YOUR_TRUSTWISE_API_KEY>" } } } } 4. For Cursor, add the docker command with arguments run, -i, --rm, -e, TW_API_KEY, -e, TW_BASE_URL, and image ghcr.io/trustwiseai/trustwise-mcp-server:latest to your settings. Set TW_API_KEY and optional TW_BASE_URL. 5. Restart your client to load the tools.

What you can do with Trustwise

• Evaluating the faithfulness and relevancy of RAG pipeline responses against provided source contexts and original user queries. - Detecting sensitive personally identifiable information and prompt injection risks in prompt inputs and model outputs before delivery. - Auditing conversational tone, formality, toxicity, clarity, and adherence to system instructions across generative agent interactions. - Estimating the financial costs and environmental carbon footprint associated with large language model inference requests. - Testing the stability and consistency of AI completions across repeated runs to ensure reproducible application outputs.

Key facts

• https://github.com/trustwiseai/trustwise-mcp-server
• AI & LLM Tooling
• alignment, evaluation, performance, safety, trustwise

Part of MCP Servers

How do I install Trustwise?

You can run the Trustwise server using Docker by pulling the container image ghcr.io/trustwiseai/trustwise-mcp-server:latest. In clients like Claude Desktop or Cursor, configure the client to execute the Docker run command while providing your Trustwise API key through the TW_API_KEY environment variable.

What can Trustwise do?

Trustwise provides evaluation tools that assess model output quality across multiple dimensions. It evaluates response faithfulness, relevance, summarization quality, toxicity, tone, and policy adherence. It also identifies prompt injection risks, detects personal data exposure, and calculates inference costs and carbon footprints.

Which MCP clients work with Trustwise?

Trustwise works with any Model Context Protocol client capable of launching local process commands using Docker. Tested configurations are provided in the repository for Claude Desktop and Cursor, though it functions in any client adhering to standard MCP specifications.

Is Trustwise open source?

Yes, the Trustwise MCP server is open source software released under the terms of the MIT license. The source code and configuration details can be accessed directly on its GitHub repository.

What credentials are required to use Trustwise?

You need a Trustwise API key provided via the TW_API_KEY environment variable. If you are connecting to a private or self-hosted Trustwise deployment, you can also specify the target endpoint using the optional TW_BASE_URL environment variable.