* use select() before readline
* support for setting static metadata for MCP sessions
* nest extra mcp metadata in metadata object
* unify session metadata
* extra metadata tests
* use empty object as parameters, if None
* list_tools as tool call
* offset indices in tests
* test: adjust addresses
* mcp: make error reporting configurable
* line logging
* log version
* verbose logging + loud exception failure
* add server and client name to policy get
* append trace even if not pushing
* port tools/list message support to SSE
* use python -m build
* adjust guardrail failure address
* support for blocking tools/list in SSE
* use error-based failure response format by default
* tools/list test
* don't list_tools in stdio connect
* flaky test: handle second possible result in anthropic streaming case
---------
Co-authored-by: knielsen404 <kristian@invariantlabs.ai>
Luca Beurer-Kellner