Stop feeding context by hand.

With our MCP server integration, it fetches exactly what it needs, right when it needs it.

Note 1 — registration

npx chat-recall init

Finds every AI coding tool already on your machine and configures each one for you.

Every tool, grouped

Resume and cold start

recall_smart_resume

recall_wake_up

recall_recent

recall_search

recall_memory_search

recall_user_prompts

Inspect one session

recall_context

recall_summary

recall_diff

recall_commits

Knowledge graph

recall_kg_query

recall_kg_add

recall_kg_invalidate

Security

recall_security_summary

recall_security_dismiss

Removing things

recall_forget

recall_exclude_path

These tools can only narrow, never widen

No tool call ever adds to an allowlist or extends retention. Both ask you every time. Widening what's shared is something only you do, from the CLI.

63 tools, 28 registered by default

The more tools an assistant has to pick from, the worse it gets at choosing the right one, so a fresh install only registers the ones people actually reach for.

CHAT_RECALL_MCP_PROFILE=full

Concurrent agents share one engine

Every session is a lightweight 4 KB relay onto one background process per machine. In testing, that took memory per session from 87 MB down to 50 MB.

CHAT_RECALL_NO_DAEMON=1 opts out

Fig. 4 — chat-recall searchReal format, sample session

$ chat-recall search "why did we drop the pooler"

Results for: "why did we drop the pooler"

#1  prepared statements break in transaction mode
    Project: api-gateway
    Score: 71/100
    Summary: Moved the writer off the transaction pooler.
    Transaction pooling rewrites the session, so a named prepared
    statement resolves against the wrong backend.
    Resume: claude --resume 8f2c1a4e

#2  connection ceiling math for the API tier
    Project: api-gateway
    Score: 44/100
    Resume: claude --resume 1b93de77

Which kind of memory server is this?

Three unrelated products ship under the label. The three kinds of MCP memory server sets out what separates them, and when you want each one.