MCP Server
The @apex-inc/mcp-server is a Model Context Protocol server that gives AI coding agents direct access to Apex's intelligence layer. Instead of switching between your editor and the dashboard, your agent can plan experiments, record beliefs, track deployments, and evaluate features — all from within your IDE.
Why Use the MCP Server?
Traditional A/B testing requires context-switching: open the dashboard, configure an experiment, go back to the code, wire up variants, re-check results. The MCP server collapses this into a single workflow inside your editor.
- Plan before you build —
plan_experimentandevaluate_featureanalyze your belief graph and past experiments before you write a line of code - Decision guardrails —
predict_impactsearches historical data for similar changes and warns you about likely outcomes - Full experiment lifecycle — create, activate, pause, archive, and promote experiments without leaving your IDE
- Identity and event tracking — instrument events and identify users from agent workflows, CI pipelines, or IDE actions
- Read-only context — MCP resources like
apex://beliefsandapex://experiments/activegive agents ambient awareness of your growth state
Capabilities
Tools (23)
Tools are actions the agent can take. They're organized into four categories:
| Category | Tools | Purpose |
|---|---|---|
| Intelligence | plan_experiment, start_reasoning, create_belief, suggest_experiment, predict_impact, evaluate_feature, log_prediction, record_outcome | Belief management, impact prediction, pre-build evaluation |
| Experiments | create_experiment, activate_experiment, pause_experiment, archive_experiment, list_experiments, get_results, promote_winner | Full experiment lifecycle |
| Tracking | track_event, identify_user, track_deployment, check_deployment | Event ingestion, identity stitching, deployment monitoring |
| Context | list_orgs, switch_org, list_projects, switch_project | Multi-org switching, project selection |
Resources (6)
Resources provide read-only context that agents can inspect at any time:
| Resource | Description |
|---|---|
apex://beliefs | All recorded beliefs with confidence scores |
apex://experiments/active | Currently running experiments |
apex://ledger | Decision log — every belief update with reasoning |
apex://intelligence-score | Your team's overall intelligence score |
apex://intelligence | Full intelligence breakdown by category |
apex://calibration | Prediction accuracy — how well your team forecasts outcomes |
Prompts (2)
Prompts are pre-built conversation starters:
- new-experiment — Guided flow to design and launch a new experiment
- experiment-review — Analyze results of a running or completed experiment
Next Steps
- Install the MCP server in your editor
- Browse all tools with parameter details
- Learn common workflows for experiment-driven development