Install the OpenTelemetry Collector (core or contrib distribution) for your platform and create a collector config YAML file
Define one or more receivers under the 'receivers:' key (e.g. otlp with grpc/http endpoints, hostmetrics, or filelog) supplying listen address and any protocol-specific options
Add processors under 'processors:' such as 'batch' (to reduce export calls), 'memory_limiter' (to cap RAM usage), and 'resourcedetection' (to enrich resource attributes)
Configure exporters under 'exporters:' pointing to your backend (e.g. otlp/http with endpoint, headers; or prometheusremotewrite with endpoint)
Wire components together in a named pipeline under 'service.pipelines' for each signal type (traces, metrics, logs), listing receivers, processors, and exporters in order
Start the collector with '--config collector-config.yaml' and validate via the collector's internal health-check endpoint and zpages extension
Known gotchas
Processors listed in a pipeline are applied in declaration order; placing 'memory_limiter' first is strongly recommended to prevent OOM before batching
The 'batch' processor can introduce latency if 'send_batch_max_size' and 'timeout' are not tuned; default timeout is 200ms which may be too low for high-throughput workloads
Contrib receivers and exporters are not included in the core binary; you must use the contrib distribution or build a custom collector with ocb (OpenTelemetry Collector Builder)
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp