Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xantorres/engram/llms.txt

Use this file to discover all available pages before exploring further.

These three commands handle the operational layer of Engram: moving memories from the staging queue into active recall, connecting harnesses to Engram over the Model Context Protocol, and running the MCP server itself. sync is safe to run repeatedly — it defaults to a dry-run so you can preview changes before committing. init is a one-time wiring step per harness. serve keeps the MCP server alive for agent sessions that communicate over stdio.

​
sync

Runs the promotion bridge over every pending candidate and decides how each one should be routed: direct append, queued for human review, or skipped. Without --apply, the command is a harmless dry-run that shows what would happen. 
engram sync [--apply]

​
Arguments & Options

​
--apply
boolean
default:"false"
Execute the routing decisions and write changes to the memory store. Without this flag, Engram prints a dry-run summary and exits without modifying anything.Both this flag and autopromote being enabled are required for actual writes to occur. Enable autopromote via autopromote: true in ~/.config/engram/config.toml or the environment variable ENGRAM_AUTOPROMOTE=true. If autopromote is not set, --apply is accepted but no writes occur.

​
Output

The summary line reports three counters: 
[dry-run] append=<N> queue=<N> skip=<N>
Or, when --apply is active: 
append=<N> queue=<N> skip=<N>
Each routed memory is listed individually beneath the summary: 
  append mem-0001 [tooling] I prefer pnpm over npm  (low-risk)
  queue  mem-0002 [fiscal]  My AWS budget cap is $500/mo  (tier-3)
  skip   mem-0003 [preference] I like dark mode  (duplicate)
Route meanings:
RouteMeaning
appendLow-risk fact; promoted directly to recall on --apply.
queueHigher-risk fact (tier-3); moved to the review queue, requires promote --confirm.
skipDuplicate or otherwise filtered out; no action taken.

​
Examples

Preview what sync would do without making any changes: 
engram sync

[dry-run] append=2 queue=1 skip=0
  append mem-0001 [tooling]    I prefer pnpm over npm       (low-risk)
  append mem-0005 [preference] I prefer 2-space indentation (low-risk)
  queue  mem-0002 [fiscal]     My AWS budget cap is $500/mo (tier-3)
Apply the changes (requires autopromote to be enabled): 
ENGRAM_AUTOPROMOTE=true engram sync --apply

append=2 queue=1 skip=0
  append mem-0001 [tooling]    I prefer pnpm over npm       (low-risk)
  append mem-0005 [preference] I prefer 2-space indentation (low-risk)
  queue  mem-0002 [fiscal]     My AWS budget cap is $500/mo (tier-3)
Memories routed to queue are not promoted automatically even with --apply. They require an explicit engram promote <id> --confirm to enter recall. This is intentional for higher-risk memory kinds such as fiscal, identity, and constraint.
Run engram sync (no flags) as often as you like — it is read-only in dry-run mode and will never corrupt your memory store.

​
init

Prints the JSON configuration snippet you need to add to a harness’s MCP config file in order to wire it to the Engram MCP server. The snippet is harness-specific and ready to paste. 
engram init <harness>

​
Arguments & Options

​
harness
string
required
The agent harness to generate config for.Valid values: claude-code, codex, opencode.

​
Examples

Generate a Claude Code MCP config snippet: 
engram init claude-code

{
  "mcpServers": {
    "engram": {
      "command": "engram",
      "args": ["serve"]
    }
  }
}
Generate config for Codex: 
engram init codex
Generate config for OpenCode: 
engram init opencode
Paste the output into the appropriate MCP config file for your harness. For Claude Code this is typically ~/.claude/claude_desktop_config.json. Once wired, agents can call the remember and recall MCP tools and read the memory://recall resource without any extra setup.

​
serve

Starts the Engram MCP server over stdio. The server exposes two tools (remember, recall) and one resource (memory://recall) to any connected harness. This command is normally invoked automatically by the harness using the config produced by engram init, but you can also run it directly for debugging. 
engram serve

​
Arguments & Options

This command takes no arguments or options.

​
Exposed MCP surface

TypeNameDescription
ToolrememberStage a fact into memory. Mirrors engram remember.
ToolrecallQuery promoted memories. Mirrors engram recall.
Resourcememory://recallStatic snapshot of all promoted memories for context injection.

​
Example

engram serve

engram MCP server running (stdio)
serve blocks and communicates over stdin/stdout. It is designed to be launched as a subprocess by your harness, not run interactively. Use engram init <harness> to get the correct launch config rather than starting it manually.
Only one engram serve process should be running per user session. Starting a second instance against the same store may cause write conflicts.

Build docs developers (and LLMs) love

Get started for freeTalk to us