Fidence in Claude

Fidence is the prediction layer. In Claude, it turns your own data into a custom predictive model, scores every record, and writes the scores back where work happens.

Add to Claude

Community connector. Sign in with Google using your Fidence account.

Connect

In Claude, open Settings, then Connectors.
Add Fidence from the Connectors Directory, or add a custom connector with this URL:

https://mcp.fidence-ai.com/mcp

A worked example: Monday-morning churn triage

A customer success lead keeps their customer list on a monday.com board. This is the whole flow, in their words.

You say	What happens
Connect Fidence.	Sign in with Google once. Claude confirms which account it uses.
What boards do I have in monday?	Your boards with item counts.
Build a churn model from my customers board. Churned customers are marked in Exited.	Fidence imports the labelled rows to learn from and the open rows to score, then starts training.
(Claude checks progress for you.)	Training finishes in about 5 to 10 minutes.
How good is it, and what drives churn?	Model quality (for example ROC AUC 0.88, precision 0.75) and the top drivers.
Who are the top 5 at risk?	Ranked customers with churn probability and risk tier.
What do the high-risk ones have in common, and what should my team do this week?	Fidence's analysis plus suggested follow-up questions.
Put the scores on the board.	A preview first, then the scores are written as new columns after you confirm.
(No Claude needed.)	Your board automations and your own agents act on the new risk column.

Tools

Claude asks before running any tool that writes. Reading tools run without a prompt once you allow them.

Tool	What it does	Access
Who am I `whoami`	Which Fidence account the connection uses (email masked).	Reads only
List use-case templates `list_use_cases`	Ready templates: churn, lead scoring, lifetime value, fraud.	Reads only
List datasets `list_datasets`	Training and scoring datasets in your workspace.	Reads only
Run a use-case template `run_template`	Train a model from a template on its sample data.	Creates
Train a model `start_training`	Train a model on your own dataset and target column.	Creates
Training status `training_status`	Progress of a training run (5 to 10 minutes).	Reads only
Ask Fidence `ask_fidence`	Free-text questions answered by Fidence's own analysis.	Creates
Ask Fidence: status `ask_status`	The answer to a longer Ask Fidence question.	Reads only
List data sources `list_sources`	Which of monday.com, HubSpot, Salesforce, Snowflake and Databricks are connected.	Reads only
Browse a data source `browse_source`	Tables, contacts or boards you can import. Names and counts only.	Reads only
Import from a data source `import_from_source`	Bring a table, contact list or board in as a dataset to train on or score.	Creates
List monday.com boards `list_monday_boards`	Boards in your connected monday.com account.	Reads only
Import a monday.com board `import_from_monday`	Bring a board in as a dataset to train on or score.	Creates
Preview write-back `preview_writeback`	Show what write-back would change in the source. Changes nothing.	Reads only
Write scores back to the source `write_back`	Write prediction columns back where the data came from, after you confirm.	Writes after you confirm
List models `list_models`	Your models with their headline metrics.	Reads only
Describe a model `describe_model`	Full metrics, features and what drives the predictions.	Reads only
List prediction results `list_results`	Prediction runs, newest first.	Reads only
Get prediction scores `get_scores`	Ranked scores with risk tiers, pseudonymous ids by default.	Reads only

Your data

Every model is trained only on your data and stays in your own workspace.
Scores show pseudonymous ids by default. Full rows are returned only when you confirm.
Fidence receives the tool calls, never your conversation. How the connector handles data

Remove

Disconnect Fidence in Claude's connector settings. Access stops immediately.

Help

Write to support@fidence-ai.com.