Support Matrix
This matrix separates the MCP core contract from client-specific behavior. The
core server returns full payloads in structuredContent; each client decides how
much of that payload is exposed to the model in its own UI/runtime.
| Area |
Status |
Notes |
| Windows local development |
Supported |
Primary practical target today. |
| Linux CI |
Supported |
Covered in GitHub Actions. |
| Windows CI |
Supported |
Covered in GitHub Actions. |
| macOS local development |
Expected |
No CI coverage yet. |
| Public support level |
Supported local-first core |
The 1.0.x line has no hosted service or formal SLA. |
| Local SQLite persistence |
Supported |
Plaintext by default; treat the DB as sensitive project data. |
| MCP stdio retrieval |
Supported |
Current primary retrieval surface. |
| Optional local daemon |
Supported |
Loopback-only in 1.0.x; remote daemon bind is intentionally rejected. |
| Optional daemon bearer token |
Supported |
Protects local /mcp; this is not hosted auth, OAuth, or TLS. |
Daemon /health |
Supported |
Sanitized runtime metadata; use mem_health_runtime through MCP for detailed state. |
| FastAPI inspection API/UI |
Supported |
Local inspection at /ui. |
| Prompt-injection hardening |
Basic guardrail |
Retrieved memory is marked as advisory project context, not higher-priority instruction. |
| Local SQLite database encryption |
Not built in |
Use OS/user-profile or volume encryption if needed. |
Client/runtime evidence
Evidence levels:
L0: code/package contract only.
L1: automated fixture or local smoke.
L2: real local MCP server/daemon path exercised outside pure unit tests.
L3-R: named client/runtime made live retrieval or diagnostic MCP tool calls with mutability mode declared.
L3-W: named client/runtime demonstrated writable continuity with read_only=false.
L4: external protocol-level compatibility review without live model tool-call validation.
| Client/runtime |
Recommended mode |
Evidence level |
v1.0.1 result |
Notes |
| Codex CLI / Codex Desktop |
response_mode=compact |
L3-R plus L3-W bridge smoke and L1 fixtures |
Validated |
Compact text plus full structuredContent is usable; writable continuity is covered by bridge smoke evidence. |
| Claude Code |
response_mode=compact |
L3-R plus L3-W bridge smoke |
Validated |
Live retrieval calls observed object-root list payloads and context packs; writable continuity is covered by bridge smoke evidence. |
| Google Gemini CLI |
response_mode=verbose when compact only exposes summaries |
L3-R plus L3-W bridge smoke |
Validated with configuration caveat |
The core MCP payload is valid, but Gemini may not expose useful structuredContent to the model in compact mode. Use verbose if the model only sees a one-line summary. |
| Google Antigravity |
Same guidance as Google Gemini CLI when using the same MCP bridge behavior |
L0 unless directly validated |
Configuration-dependent |
Document separately if validated directly in a release run. |
| Qwen Code / Ollama workflow |
Client-appropriate MCP mode |
L3-R |
Validated as retrieval-only workflow |
This is not a native Ollama adapter; it is an MCP-capable local workflow. |
| DeepSeek / Minimax via Ollama Cloud workflow |
Client-appropriate MCP mode |
L3-R |
Validated as retrieval-only workflow |
This is not a native hosted bridge in the public core. |
| Kimi Code CLI |
Client-appropriate MCP mode |
L2 connection validation |
Connection documented |
Treat as MCP-client connection validation, not a separate backend feature or full model-run claim. |
Public evidence notes for validated rows live under
docs/validation/evidence. Evidence files
summarize the observed client/runtime, mode, tools, limits, and caveats without
publishing private transcripts or operational database contents.
MCP response behavior
| Surface |
Status |
Notes |
structuredContent |
Canonical payload |
Full tool result for MCP clients that expose structured payloads. |
List results in structuredContent |
Object-root contract |
Lists are wrapped as {items, count} for stricter MCP clients. |
| Session list results |
Object-root contract |
mem_session_list returns {items, count} for recent persisted sessions in one project. |
| Startup scope resolution |
Object-root contract |
mem_scope_resolve and mem_bootstrap_context return candidate lanes and defensive next actions before a broad pack is treated as active context. |
content.text compact |
Supported default |
Short human-readable summary for low-token clients. |
content.text balanced |
Supported |
Summary plus notice that complete data is in structuredContent. |
content.text verbose |
Supported |
JSON text view for clients that do not expose structuredContent usefully. |
pack_hash / known_pack_hash |
Supported |
Enables unchanged-pack not_modified responses. |
session_id retrieval filter |
Supported |
Optional retrieval filter for persisted memory inside a project_key; not live current-turn awareness. |
| Feature |
Status |
Notes |
mem_context_pack compact retrieval |
Supported |
Returns compressed continuity pack and approximate token budget. |
mem_session_list |
Supported |
Lists recent persisted sessions/chats so broad project scopes can be narrowed. |
mem_scope_resolve |
Supported |
Read-only resolver for explicit thread/path/query hints; does not infer live current-turn state. |
mem_bootstrap_context |
Supported |
Read-only startup helper that returns a scoped pack only when session_id is explicit, otherwise returns candidate lanes and narrowing guidance. |
mem_context_pack auto budget |
Supported |
Selects micro, normal, or full. |
mem_recent_changes |
Supported |
Summarizes deltas since the last meaningful continuity baseline. |
mem_scope_guard |
Supported |
Exposes must-not-drop scope, constraints, and closure conflicts. |
mem_open_work / mem_completion_check |
Supported |
Deterministic closure control. |
mem_provenance |
Supported |
Shows source turn/session context and payload hash. |
mem_health |
Supported |
Project health diagnostics without mutating memory. |
mem_health_runtime |
Supported |
PID, uptime, request counts, idle timeout, and lifecycle state. |
| Snapshot list/create/restore |
Supported |
Restores derived state, not silent history mutation. |
| Policies / inheritance / repairs |
Supported |
Governed local memory control. |
| Context sync metrics |
Supported |
Pack sync/skip events exposed through API/UI. |
| Optional HTTP notify wrapper |
Supported |
Secondary local path. |
Deferred or external surfaces
| Surface |
Status |
Notes |
| Hosted web bridges |
Outside public core |
Not part of the 1.0.x local-first package. |
| ChatGPT web / Claude web hosted flows |
Outside public core |
Do not treat lab bridge experiments as public core support. |
| Vector search / embeddings |
Deferred |
Not required for v1.0.x local continuity. |
| Remote telemetry |
Not included |
No remote telemetry is required for the public core. |
Version support
| Component |
Current expectation |
| Release line |
1.0.x |
| Python |
3.12+ |
| SQLite |
Local system SQLite with FTS5 support preferred |
| Integration model |
MCP config plus local ingestion path |