Issues

MCP proxy grants full memwal:read memwal:write scope to requests with no Authorization header when mcp_oauth is unconfigured

#710

· dun999 opened

on Aug 19, 2026

`memwal_health` reports `ok` during a total write-path outage, and the write error leaks an internal `localhost:9000` address that misleads users into debugging their own machine

#709

· Tolex081 opened

on Aug 19, 2026

recall() cannot use recency-weighted ranking that already exists in the SDK — it's implemented, typed, and wired, but only reachable through recallManual()

#708

· Collinxweb opened

on Aug 19, 2026

POST /api/mcp accepts any 64-hex Bearer — full tool session with no on-chain check, sessions pin ip_active_cap until DELETE

#707

· dun999 opened

on Aug 19, 2026

No memwal_list_namespaces tool - namespace discovery is impossible without prior knowledge

#698

· zazadra opened

on Aug 19, 2026

No memwal_delete or supersede tool via MCP - stored memories cannot be corrected

#697

· zazadra opened

on Aug 19, 2026

401 authentication error returns '<no message>' - not actionable for developers

#696

· zazadra opened

on Aug 19, 2026

memwal_recall lacks a sort_by=recency option - makes session-handoff patterns unreliable

#695

· zazadra opened

on Aug 19, 2026

memwal_recall returns duplicate records for the same blob

#694

· zazadra opened

on Aug 19, 2026

recall() has a hardcoded, unconfigurable 15-second network timeout; RecallResult.total has no documentation and its actual meaning is easy to misread

#692

· Collinxweb opened

on Aug 18, 2026

waitForRememberJob (single) throws on timeout, while waitForRememberJobs (bulk) gracefully returns for the identical failure condition

#691

· Collinxweb opened

on Aug 18, 2026

Inconsistent error typing — one typed error class exists, but only covers a single HTTP status

#687

· Collinxweb opened

on Aug 18, 2026

Navigation Menu

Uh oh!

All issues

MCP proxy grants full memwal:read memwal:write scope to requests with no Authorization header when mcp_oauth is unconfigured

`memwal_health` reports `ok` during a total write-path outage, and the write error leaks an internal `localhost:9000` address that misleads users into debugging their own machine

recall() cannot use recency-weighted ranking that already exists in the SDK — it's implemented, typed, and wired, but only reachable through recallManual()

POST /api/mcp accepts any 64-hex Bearer — full tool session with no on-chain check, sessions pin ip_active_cap until DELETE

No memwal_list_namespaces tool - namespace discovery is impossible without prior knowledge

No memwal_delete or supersede tool via MCP - stored memories cannot be corrected

401 authentication error returns '<no message>' - not actionable for developers

memwal_recall lacks a sort_by=recency option - makes session-handoff patterns unreliable

memwal_recall returns duplicate records for the same blob

recall() has a hardcoded, unconfigurable 15-second network timeout; RecallResult.total has no documentation and its actual meaning is easy to misread

waitForRememberJob (single) throws on timeout, while waitForRememberJobs (bulk) gracefully returns for the identical failure condition

Inconsistent error typing — one typed error class exists, but only covers a single HTTP status

Uh oh!

Search results