feat: add support for TRACELOOP_TRACE_CONTENT=false in mcp instrumentor

[milkisbad]

Feature description:

https://github.com/traceloop/openllmetry/blob/v0.60.0/packages/opentelemetry-instrumentation-mcp/opentelemetry/instrumentation/mcp/instrumentation.py does not respect TRACELOOP_TRACE_CONTENT=false (only the fast_mcp version does)

Why is this feature needed?:

to make flows that can't store prompts/answers be able to use the sdk

Note:

doing blank issue as can't see mcp instrumentor in list of components when creating issue

[ferhimedamine]

Configuring TRACELOOP_TRACE_CONTENT=false in the MCP instrumentor is important for production deployments where tool arguments and outputs may contain PII or proprietary data — you want the structural trace (call graph, latency, errors) without the content.

One thing worth considering for the MCP case specifically: some MCP tool outputs are more sensitive than others. A tiered approach (trace content for read-only tools, suppress for write/external-call tools) would give better observability than an all-or-nothing flag.

Separately, for teams thinking about what to store from agent traces beyond observability, dakera-deploy's memory API accepts importance-scored records — so you can store distilled insights from traces (what worked, what failed, what the agent learned) without retaining the full sensitive content.