mneme-hq 0.9.2

Complete Decision MCP input-schema documentation.

This patch release improves agent reliability by documenting every exposed Decision MCP input parameter and nested candidate/provenance field. It does not change the six-tool capability surface or any runtime authority, retrieval, proposal, enforcement, or Architecture Audit behavior.

Improved

• Added JSON Schema descriptions for every parameter on the six frozen tools:
• decision.propose
• decision.propose_batch
• decision.get
• decision.search
• decision.applicable_to
• decision.trace
• Documented required candidate content and source provenance, plus the deterministic precedence of candidate provenance over batch shared_provenance.
• Documented decision.search query matching, exact metadata filters, domain-specific filter application, and AND combination semantics.
• Added explicit tool-selection guidance for single versus batch proposals and for get, search, and trace reads.
• Clarified that empty decision.applicable_to inputs return empty match lists.
• Added live MCP tools/list assertions so parameter-description coverage is tested at the generated protocol-schema boundary.

Compatibility

• No runtime or protocol behavior changes.
• Tool names, inputs, outputs, annotations, and authority boundaries are unchanged.
• Proposal acceptance, rejection, activation, supersession, enforcement, and trusted evidence remain outside the MCP authority boundary.
• mneme.audit/v1 is unchanged.

Install

:::bash
pip install "mneme-hq[mcp]==0.9.2"

Start the local stdio server:

:::bash
mneme decision-mcp