EPOCH MCP
Your agent. A deeper research engine.
Bring quality data and deterministic research computation to your coding agent. Iterate on studies, retain the evidence and rerun saved rules as new data arrives, inside your existing workflow.
Explore the workflowExplore the edge
Bring your research brief, holdings and cost assumptions. Your agent uses Epoch data and compute to test the study, then inspects the results.
I’ll read the local brief and costs, use Epoch for the shared data and deterministic runs, then review the returned evidence before revising the study.
Used Epoch · Read brief.md and costs.csvWorking
Read the project brief and cost file. Resolve the liquid index futures universe against Epoch data and freeze the development, validation and final holdout windows.
Used Epoch · Executed the baseline and two alternativesQueued
Built a fast breakout baseline, a trend-filtered breakout and a time-series momentum rule. Each uses the same data window and cost assumptions.
Used Epoch · Inspected failures and reran bounded trialsQueued
Inspect the returned trial table before changing lookback or filter settings. Run a bounded set of 48 declared combinations on development data; retain failed candidates and leave the final holdout untouched.
Used Epoch · Checked unseen periods and trading costsQueued
Shortlisted on validation evidence, then evaluated the frozen candidate on the final holdout. The fast breakout lost much of its apparent advantage after costs.
Used Epoch · Ran Monte Carlo robustness analysisQueued
Resampled backtest returns to examine the spread of outcomes and drawdowns under the simulation assumptions. This estimates uncertainty; it does not establish a future return.
Used Epoch · Saved the candidate and its evidenceQueued
Saved the rule, parameters, data window and evaluation history together. Selected the trend-filtered breakout within the example drawdown and turnover constraints.
Test what holds
Inspect the trade-offs, validation results and robustness checks. Keep the selected rule and its evidence together.
Used Epoch · Compared rules, tested parameters, challenged the resultsThe strongest backtest
wasn’t the strongest candidate.
The fast breakout led in training, then weakened after costs and on unseen data. The trend-filtered rule held up better within the drawdown and turnover limits.
| Strategy | Train Sharpe | Validation Sharpe | Max drawdown | Decision |
|---|---|---|---|---|
| Fast breakout | 2.34 | 0.66 | −24.8% | Cost-sensitive |
| Trend-filtered breakout | 1.86 | 1.42 | −12.6% | Selected |
| Time-series momentum | 1.57 | 1.10 | −16.9% | Retained |
Trend-filtered breakout. Its settings, data window and full trial history stay attached to the strategy.
Keep it in play
Monitor the saved strategy. A new market event brings you back to the same rules, with a fresh signal or target allocation.
The saved method and risk limits stay fixed. On each scheduled run, the agent calls Epoch with fresh inputs, inspects the response and posts changes with their evidence.
Trend-filtered breakout
- Evaluate
- After each market close
- Notify on
- New long signal · target weight change
- Keep fixed
- Saved rules, parameters and risk limits