F
Featrix

Talk to your data through Claude and ChatGPT

This is Featrix's Model Context Protocol server — connect it to Claude Desktop, claude.ai, Claude Code, or ChatGPT and train models, run predictions, and explore your data directly from a conversation. 14 tools, no Python required — every call runs as you, using your own Featrix API key.

Connect

Get an API key from your Featrix account, then add this server to your MCP client.

Claude Desktop / claude.ai — mcpServers config
{
  "mcpServers": {
    "featrix": {
      "url": "https://mcp.featrix.com/api/mcp",
      "headers": {
        "Authorization": "Bearer fx_your_api_key_here"
      }
    }
  }
}
Claude Code — CLI
claude mcp add --scope project --transport http featrix https://mcp.featrix.com/api/mcp \
  --header "Authorization: Bearer ${FEATRIX_API_KEY}"

For stdio-only clients, bridge through mcp-remote: npx mcp-remote https://mcp.featrix.com/api/mcp --header "Authorization:Bearer fx_your_api_key_here"

No key? Discovery (listing tools) works without one — calling a tool that touches real data returns a clear error telling you to connect one, instead of running as someone else's account.

14 tools

Ingest & Train
list_modelsList all trained foundational models
train_selfsupervised_foundation_no_labels_neededTrain on your data — no labels needed
check_training_statusCheck training progress
train_predictor_for_specific_targetFine-tune a predictor for one column
Predict & Impute
predict_any_columnPredict any column for a record
fill_in_missing_valuesFill in missing fields in a partial record
batch_predictPredict across multiple records at once
Explore
find_similar_recordsNearest neighbors in embedding space
compare_what_if_scenariosSee how predictions change under different scenarios
encode_recordsTransform records into embedding vectors
Explain & Feedback
explain_what_drives_predictionFeature importance via leave-one-out ablation
describe_what_model_knowsColumns, types, and capabilities a model learned
suggest_questionsInteresting questions to ask about your data
send_ground_truth_feedbackReport the actual outcome for a prediction