Reexpress

- First reliable, statistically robust AI second opinion for AI workflows
- Computes predictive uncertainty using an SDM estimator locally
- Dynamically updatable via ReexpressAddTrue and ReexpressAddFalse tools
- Simple, conservative file access system
- Runs on macOS and Linux with minimal compute requirements
- Enables reasoning with verification for tool-calling LLMs

For Tool Calling Llms E G Claude Opus 4 7 And Mcp Clients Running On Macos Tahoe 26 Or Later On Apple Silicon Or Linux

Reexpress MCP Server is a drop-in solution to add state-of-the-art statistical verification to your complex LLM pipelines, as well as your everyday use of LLMs for search and QA forsoftware development and data science settings. It's the first reliable, statistically robust AI second opinion for your AI workflows.

Simply install the MCP server and then add the Reexpress prompt to the end of your chat text. The tool-calling LLM (e.g., Anthropic's LLM model Claude Opus 4.7) will then check its response with the provided pre-trained ReexpressSimilarity-Distance-Magnitude (SDM) estimator, which ensembles gpt-5.5-2026-04-23, gemini-3.1-pro-preview, and gemini-embedding-2, along with the output from the tool-calling LLM, and calculates a robust estimate of the predictive uncertainty against a database of training and calibration examples from the OpenVerification1 dataset. Unique to the Reexpress method, you can easily adapt the model to your tasks: Simply call the ReexpressAddTrue or ReexpressAddFalse tools after a verification has completed, and then future calls to the Reexpress tool will dynamically take your updates into consideration when calculating the verification probability. We also include the training scripts for the model, so that you can run a full retraining when more substantive changes are needed, or you want to use alternative underlying LLMs.

[!NOTE] In addition to providing you (the user) with a principled estimate of confidence in the output given your instructions, the tool-calling LLM itself can use the verification output to progressively refine its answer, determine if it needs additional outside resources or tools, or has reached an impasse and needs to ask you for further clarification or information. That's what we callreasoning with SDM verification--- an entirely new capability in the AI toolkit that we think will open up a much broader range of use-cases for LLMs and LLM agents, for both individuals and enterprises.

Data is only sent via standard LLM API calls to Azure/OpenAI and Google, with the gemini-3.1-pro-preview calls given standard web search access through the API; all of the processing for the SDM estimator is done locally on your computer. Reexpress MCP has a simple and conservative, but effective, file access system: You control which additional files (if any) get sent to the LLM APIs by explicitly specifying files via the file-access tools ReexpressDirectorySet() and ReexpressFileSet().

Version 2.4.0 uses gpt-5.5-2026-04-23 and gemini-3.1-pro-preview as the generative models. As with 2.3.0.preview, gemini-embedding-2 replaces the local granite-3.3-8b-instruct model as the agreement representation model. This greatly simplifies running the Server, since you no longer need to locally run a multi-billion parameter model. Additionally, we have also expanded the OpenVerification1 dataset with new examples. See themodel cardfor details.

[!TIP] The Reexpress MCP server is straightforward to setup relative to other MCP servers, but we assume some familiarity with LLMs, MCP, and command-line tools. Our target audience is developers and data scientists. Only add other MCP servers from sources that you trust, and keep in mind that other MCP tools could alter the behavior of our MCP server in unexpected ways.