Skip to content

mcp_capability_matrix.md

Latest commit

 

History

History
98 lines (78 loc) · 4.93 KB

mcp_capability_matrix.md

File metadata and controls

98 lines (78 loc) · 4.93 KB

ZCP MCP Capability Matrix

This matrix tracks parity against the official MCP specification repository and the official MCP Python SDK from the perspective of this workspace.

ZCP serves two surfaces from one runtime:

  • the MCP-compatible surface for interoperability
  • the native ZCP surface for lower token usage and tighter orchestration

The table separates three questions:

  • is the feature implemented
  • is there compatibility coverage here
  • what does the native ZCP path improve
MCP Feature Implemented On MCP Surface Compatibility Coverage Here ZCP Native Advantage
Lifecycle: initialize, initialized, ping Yes Yes Smaller native envelopes
Tools: tools/list, tools/call Yes Yes Compact registry, handle-first results, and semantic workflow profiles
Tool metadata: title, annotations, outputSchema, execution, _meta Yes Yes Canonical metadata can stay out of prompt context
Resources: resources/list, resources/read, subscribe/update flow Yes Yes for core flows Large content can stay off-context
Resource templates: resources/templates/list Yes Yes Compact template discovery
Prompts: prompts/list, prompts/get Yes Yes Expanded prompt artifacts do not need to remain model-visible
Completion: completion/complete Yes Yes Native refs are smaller
Logging: logging/setLevel, notifications/message Yes Yes in SDK/runtime coverage Operational summaries can stay outside prompt context
Roots: roots/list, notifications/roots/list_changed Yes Yes in SDK/runtime coverage Same semantics with smaller envelopes
Progress notifications Yes Partial Native progress payloads are shorter
Sampling Yes Partial Less repeated schema/context echo
Elicitation Yes Partial Compact accept/decline flow summaries
Tasks basic methods and statuses Yes Partial Durable off-context state and result storage
Stdio transport Yes Yes Same runtime with shorter native traffic
Streamable HTTP on /mcp Yes Yes Native /zcp remains smaller
WebSocket transport on /ws Yes Yes for core client flow Native socket traffic can be more compact
Authorization metadata Yes Partial Native auth profile metadata stays runtime-owned
OAuth code flow, PKCE, refresh, registration, revocation Yes Partial Same provider model can be reused for native surfaces

How To Read This Matrix

  • Yes in implementation means the feature is present in the current runtime.
  • Yes in compatibility coverage means this repository has concrete validation for the feature or transport against the current implementation.
  • Partial means the feature exists and works in core scenarios, but broader interoperability or edge-case coverage is still being tracked.

Current Position

The current implementation now includes:

  • stdio interoperability
  • streamable HTTP interoperability on /mcp
  • websocket interoperability on /ws
  • OAuth metadata and protected resource metadata
  • OAuth authorization code flow with PKCE
  • refresh token exchange
  • dynamic registration and revocation
  • task-augmented tool calls
  • async task lifecycle states

The strict remaining work is mostly:

  • broader compatibility coverage for progress, sampling, elicitation, and tasks
  • deeper reconnect and soak coverage for long-lived transports
  • broader auth interoperability coverage beyond the current validated flows

For the precise remaining list, use mcp_gap_todo.md.

Verified Performance Snapshot (Excel Semantic Suite v5)

Run metadata:

  • date: 2026-03-17
  • model: deepseek-chat (https://api.deepseek.com)
  • repeats: 1
  • cases: 37 (Tier A/B/C/D)
  • compared backends:
    • zcp_client_to_native_zcp
    • mcp_client_to_zcp_mcp_surface
  • artifacts:
    • zero-context-protocol-python/benchmark_reports/full_semantic_compare_v5/excel_llm_token_benchmark.json
    • zero-context-protocol-python/benchmark_reports/full_semantic_compare_v5/excel_llm_token_benchmark.md
    • zero-context-protocol-python/benchmark_reports/full_semantic_compare_v5/semantic_benchmark_summary.md
Scope Native ZCP Avg Total MCP Surface Avg Total MCP/Native Ratio Native Quality (Answer/Workbook/Tool)
Overall 8027.9 30723.7 3.83x 100.0% / 97.3% / 100.0%
Tier A 15979.4 17613.2 1.10x 100.0% / 93.8% / 100.0%
Tier B 1826.6 29239.4 16.01x 100.0% / 100.0% / 100.0%
Tier C 2091.1 72113.9 34.49x 100.0% / 100.0% / 100.0%
Tier D 2018.3 19375.7 9.60x 100.0% / 100.0% / 100.0%

Interpretation notes:

  • native ZCP wins average token usage in every tier for this run
  • native ZCP is lower-token in 25/37 cases; the remaining 12 are all Tier A primitive cases
  • quality gaps are largest in complex tiers on the MCP surface path:
    • Tier C tool compliance: 57.1% vs native 100.0%
    • Tier D tool compliance: 16.7% vs native 100.0%