Reduce MCP tool surface area

[cl0ckt0wer]

Problem

agentmemory currently exposes too many MCP tools, which makes the tool list noisy for agents and increases the chance that models choose the wrong entry point. This came up while refreshing PR #279, where the branch had to reconcile older docs that referenced 51/53 tools with the newer consolidated tool surface.

PR: #279

Why this matters

• Large MCP tool lists create selection overhead for coding agents.
• Similar legacy tools and consolidated tools compete with each other.
• Tool-count docs/manifests drift easily when the surface changes.
• Users see a more complex integration story than necessary.

Proposal

Keep the public/default MCP surface intentionally small and operation-driven:

• Prefer consolidated tools such as memory_search, memory_store, lesson, admin, task, and slot.
• Hide legacy one-off aliases by default, or only expose them behind an explicit compatibility mode.
• Make AGENTMEMORY_TOOLS=core|all semantics clear and test-covered.
• Keep README, plugin manifests, generated metadata, and standalone MCP tests tied to a single source of truth for tool counts.

Acceptance criteria

• Default MCP tool list is small enough for agents to scan reliably.
• Legacy aliases remain callable through compatibility/proxy paths where needed, but are not all advertised by default.
• Tool counts are generated or checked in tests so docs/manifests cannot drift.
• PR feat: auto-create lessons in flow-compress, consolidate, and new MCP tools #279 can reference this issue as the follow-up for broader tool-surface cleanup.