openinterp-mcp

Run mechanistic interpretability experiments and probe model features on your own compute.

Source

third_party_registry

Updated

2026-07-13

// securityCaution

Runs locally
Open source

Running security audit…

Credentials
Network egress
Code execution
Data access
Supply chain

// install

Copy the install command and let the AI configure it · recommended for beginners

No copy-paste install info for "openinterp-mcp" yet — see the docs or source repo.

// examples

Run an SAE feature experiment

Prompt

Use openinterp-mcp in my Colab environment to design and run an SAE feature experiment on this language model layer, compare feature activations across the provided text samples, and return key findings with next-step recommendations.

Expected

A concise experiment summary with feature activation comparisons, notable patterns, conclusions, and suggested follow-up experiments.

Perform probe-causality analysis

Prompt

Using openinterp-mcp, run a probe-causality analysis on the specified neurons or representation subspace, test their relationship to model output behavior, and summarize the procedure, metrics, and conclusions.

Expected

A structured report showing which representations relate to the target behavior and how causal interventions affect outputs.

Plan an interpretability research workflow

Prompt

I want to study this model’s internal mechanisms for a specific task. Use openinterp-mcp to plan an executable research workflow, including experiment order, required tools, data preparation, and a results logging template.

Expected

A clear research plan that helps the user run mechanistic interpretability experiments step by step and keep reproducible records.