Ask the pipeline.
Change it with tools.
TrackLayer Copilot runs in /copilot and through npx @tracklayer/mcp-server. The agent exposes 30 MCP tools for events, deliveries, audiences, taxonomy, insights, destinations, replay, and alert operations.
Built from the product surface, not brochure claims.
What the team actually gets.
Evidence before answers
The agent reads event rows, delivery attempts, detector evidence, and taxonomy versions before summarizing.
Operator actions
Replay failures, create audiences, inspect schema drift, and draft destination fixes from the same interface.
Dashboard-native
The /copilot route sits beside event stream, insights, deliveries, and audiences so answers link back to records.
CLI-ready
Agencies can run the MCP server locally and connect TrackLayer to their preferred agent client.
How it compares to ordinary tracking work.
- Event investigation
- tool-backed
- Delivery remediation
- replay tool
- Audience creation
- agent action
- Schema drift explanation
- taxonomy-aware
- Local agent support
- MCP server
- Event investigation
- manual SQL
- Delivery remediation
- manual retry
- Audience creation
- manual segment
- Schema drift explanation
- manual QA
- Local agent support
- not public
- Event investigation
- dashboard filter
- Delivery remediation
- support ticket
- Audience creation
- downstream only
- Schema drift explanation
- not native
- Local agent support
- not public
Real merchant-shaped cases and measurable signals.
The references an operator can inspect.
Where this matters in production.
Account managers needed to explain delivery drops across 14 merchants.
Copilot grouped failures by root cause and produced replay actions per merchant.
Paid media asked for a high-intent audience before the first campaign review.
The MCP audience tool compiled add_to_cart and checkout_started users with consent filters.
Engineering renamed a lead field and broke LinkedIn match quality.
Copilot tied the regression to schema drift evidence and the affected canonical event.