Reexpress Model Context Protocol Server

What is Reexpress Model Context Protocol Server about?

Reexpress Model Context Protocol Server adds a statistical second‑opinion layer to LLM workflows. It uses a pre‑trained Similarity‑Distance‑Magnitude (SDM) estimator that ensembles multiple LLM explanations (gpt‑4.1‑2025‑04‑14, o4‑mini‑2025‑04‑16‑high, gemini‑2.5‑pro, and a locally run granite‑3.3‑8b‑instruct) to compute a predictive uncertainty score for any response generated by Claude Opus 4.

How to use Reexpress Model Context Protocol Server?

Install the server following the steps in INSTALL.md (Linux/macOS required).
Configure API keys for Anthropic, Azure/OpenAI, and Google in the provided JSON settings (code/reexpress/mcp_settings.json).
Add the Reexpress prompt to the end of each chat interaction. The server will invoke the SDM estimator, return an XML‑wrapped verification block, and optionally let Claude refine its answer based on the verification result.
Interact with tools such as ReexpressAddTrue, ReexpressAddFalse, reexpress_add_ood(), ReexpressDirectorySet(), and ReexpressFileSet() to update the support set, manage file access, and guide future verification.

Key features of Reexpress Model Context Protocol Server

Ensemble verification across four cutting‑edge LLMs, with local inference for the embedding model.
Dynamic adaptation via ReexpressAddTrue/False and out‑of‑distribution support tools.
Static HTML introspection of nearest training examples for each prediction.
Version 1.1.0 upgrades: added gemini‑2.5‑pro, higher reasoning budget for o4‑mini, local granite embedding, confidence bins (>=90%, <90% (use with caution), Out‑of‑distribution).
Reduced confidence threshold (α' = 0.9) to admit ~62 % of in‑distribution cases with ~92 % marginal accuracy.
Comprehensive documentation (HOW_TO_USE, GUIDELINES, FAQ, EVAL, DATA, etc.) and training scripts for full re‑training.

Use cases of Reexpress Model Context Protocol Server

Software development assistants that need reliable verification of code suggestions or debugging output.
Data‑science analysis pipelines where accuracy of generated explanations or statistical insights must be quantified.
Enterprise AI agents requiring a quantified second opinion before taking autonomous actions.
Research prototyping that benefits from quick adaptation of verification models to new domains via the provided training scripts.
Human‑in‑the‑loop workflows where users decide to trust or reject model outputs based on a clear confidence metric.

FAQ from the Reexpress Model Context Protocol Server

Which LLMs are used in the ensemble? gpt‑4.1‑2025‑04‑14, o4‑mini‑2025‑04‑16‑high, gemini‑2.5‑pro, and a locally run granite‑3.3‑8b‑instruct for embeddings.
Do any data leave my machine? Only standard API calls to Anthropic, Azure/OpenAI, and Google are made; all SDM computation runs locally.
How can I add new training examples? Use the ReexpressAddTrue or ReexpressAddFalse tools after a verification completes; for out‑of‑distribution cases, call reexpress_add_ood().
What file‑access controls are available? Specify directories or files explicitly with ReexpressDirectorySet() and ReexpressFileSet().
Can I retrain the model? Yes, the repository includes the full training script and calibration data; follow documentation/model_details/.../train_and_eval_sdm_estimator_v1.1.0.sh.
What confidence thresholds should I use? The default bins are >=90% (trusted), <90% (use with caution), and Out‑of‑distribution for unreliable cases.
Is the server compatible with Windows? No, it currently supports only Linux and macOS.

Reexpress Model-Context-Protocol (MCP) Server

For Claude Opus 4 and MCP clients running on Linux or macOS (Sequoia 15 on Apple silicon)

Video overview¹: Here

Screenshot image of the rendered HTML output from the Reexpress tool.

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 for software 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. Anthropic's LLM model Claude Opus 4 will then check its response with the provided pre-trained Reexpress Similarity-Distance-Magnitude (SDM) estimator, which ensembles gpt-4.1-2025-04-14, o4-mini-2025-04-16-high, gemini-2.5-pro, and granite-3.3-8b-instruct (run locally), along with the output from Claude, 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, Claude 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 call reasoning 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 Anthropic, Azure/OpenAI, and Google; all of the processing for the SDM estimator is done locally on your computer. (Optionally, we highly recommend providing access to web search via your MCP client, such as via Claude Desktop or a web-search MCP server, or for closed-domain settings, access to domain-specific retrieval.) 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().

What's new in version 1.1.0

Version 1.1.0 adds a number of new capabilities:

We added gemini-2.5-pro as part of the model ensemble.
We increased the o4-mini-2025-04-16 reasoning budget from medium to high.
We replaced the API calls to the text-embedding-3-large embeddings model with the locally run ibm-granite/granite-3.3-8b-instruct model, which we use to construct the representation space over the model explanations from gpt-4.1-2025-04-14, o4-mini-2025-04-16-high, and gemini-2.5-pro.
We added the ability to introspect the predictions against the training set. You can now view the nearest match to each test instance via a static webpage that you can (optionally) generate for each prediction. This also makes it easy to quickly check how the verification estimation was determined without having to call the ReexpressView tool. See documentation/OUTPUT_HTML.md for examples.
We include the training script for the model and the model evaluation outputs over the OpenVerification1 dataset in the model directory (see the Release archive). A summary of the evaluation is available at documentation/EVAL.md.
The training and calibration data are a subset of the full OpenVerification1 dataset, which we have made available on HuggingFace datasets.
We have updated the output to the MCP server to have all content returned within XML tags to simplify use out-of-the-box for downstream, test-time search graphs. We have also updated our recommended base tool-call prompt with the following final sentence: Consider your final answer verified if <successfully_verified> True </successfully_verified> and <confidence> >= 90% </confidence>.
We have simplified the presentation of the verification confidence (i.e., the probability estimated for the binary classification prediction) in the main output to the following three bins to reflect the resolution at which we recommend using the tool:
- >= 90%
- < 90% (use with caution)
- Out-of-distribution (unreliable)
Note that we have reduced the probability threshold to 0.9 (i.e., alpha'=0.9, down from the more stringent 0.95 in version 1.0.0) to better reflect the capabilities of the current generation of models and the intended use-case of verification with a human-in-the-loop. This version admits approximately 62% of in-distribution examples at alpha'=0.9 (i.e., the proportion of valid index-conditional estimates at alpha'=0.9) from the 5k test set of the OpenVerification1 dataset, over which the marginal accuracy is approximately 92%. If you need a version with a more stringent requirement (and/or recalibration over your domain specific tasks), we provide the training code here, as noted above. For mission-critical enterprise settings and semi-autonomous agents that require alpha' > 0.9, we typically recommend training a full SDM network that composes the hidden states over all input text (prompt, response, and if applicable, the composition of the output of additional LLMs). (In contrast, the current MCP server uses an SDM estimator that marginalizes over the content of the prompt and response, and takes as input an ensemble of explanations from external LLMs. This is done to keep computational costs manageable for local deployment with existing LLM APIs.) We can assist you with building such SDM networks. Contact us!
We modified the baseline configuration in code/reexpress/mcp_settings.json.
Finally, we added a new tool function, reexpress_add_ood(), which allows you to add an out-of-distribution (label=-99) instance to the support set. (For developers and researchers, we have also updated the training and calibration routines to allow such instances to participate in training and calibration. Instances in the --ood_support_file get added to the training support for each training iteration, and thus can impact the Similarity values of training and calibration instances, if applicable.)
For researchers: Starting in commit c816516 is a script (utils_graph_output.py) to construct interactive graphs of the batch output. See the end of the training script for example use. You can click on a point to print additional information to the console. Graphs for the output of the SDM estimator in this release are saved to the repo for reference.

System Requirements

The MCP server runs on Linux and macOS. The primary requirement is that the machine running the MCP server needs to be able to locally run ibm-granite/granite-3.3-8b-instruct (via the HuggingFace transformers library). This takes as input three short (typically less than 512 tokens) model explanations and only needs to generate 1 token, so the compute requirements are relatively modest in practice.

Installation

See INSTALL.md.

[!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.

Citation

If you find this software useful, consider citing the following paper:

@misc{Schmaltz-2025-SimilarityDistanceMagnitudeUniversalVerification,
      title={Similarity-Distance-Magnitude Universal Verification}, 
      author={Allen Schmaltz},
      year={2025},
      eprint={2502.20167},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2502.20167}, 
}

The output format has changed slightly since v1.0.0 used in the video. See What's new in version 1.1.0 ↩

Reexpress Model Context Protocol Server