docs: propose standalone MCP access roadmap

[JarrettYan-519]

Summary

This PR adds an RFC for a standalone MCP access path and cross-agent Task State roadmap.

It is intentionally documentation-only and additive:

• adds a lightweight project glossary in CONTEXT.md
• adds docs/rfcs/0001-standalone-mcp-cross-agent-state.md
• proposes a Phase 1 MCP tool surface for get_context, memory_search, capture_turn, state_get, and state_update
• keeps OpenClaw and Hermes integrations unchanged and explicitly non-deprecated
• defers host-specific lifecycle hooks and direct long-term memory promotion

Motivation

The README roadmap already calls out portable memory across agents, frameworks, and devices. This RFC frames standalone MCP access as a low-risk first step toward that direction while preserving the existing OpenClaw and Hermes paths.

Validation

• Ran a docs-only Node check to verify both new files exist, the RFC glossary link resolves, and the RFC contains no TODO/TBD/FIXME or specific client names.
• Ran git diff --cached --check before committing.

No runtime tests were run because this PR changes documentation only.

[Maxwell-Code07]

Thanks for the detailed RFC! A standalone MCP access path is a forward-looking enhancement for cross-agent memory sharing.

[JarrettYan-519]

Thanks for the feedback! I’ve marked this PR as ready for review.

Since this is documentation-only and additive, would you be open to merging the RFC first as a roadmap proposal? If you prefer, I can also follow up with a separate implementation PR for the Phase 1 MCP adapter.

[YOMXXX]

Duplicate cleanup / re-scope note:

The standalone MCP / adapter-access part of this roadmap now overlaps with the verified #235 base layers:

• feat(adapters): add gateway client adapter kit #316 for the lightweight Gateway client / generic lifecycle adapter.
• feat(adapter): add Codex MCP gateway bridge #372 for the focused stdio MCP Gateway bridge.
• docs(adapters): clarify adapter PR boundaries #441 for adapter contribution boundaries and scope guidance.

This PR is docs-only and still has potentially useful RFC material, especially around cross-agent task state. I would not use it as the canonical adapter/MCP roadmap anymore, though. Recommendation: re-scope the RFC to the unique cross-agent state model and explicitly reference #316/#372/#441 as the current implementation/documentation baselines instead of proposing a parallel MCP tool surface.