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.

Wiring an agent means registering engram-mcp as a Model Context Protocol server inside that agent’s configuration file. Once wired, the agent gains a remember tool it can call to store facts and a memory://recall resource it can read at the start of any session — giving it durable, project-scoped memory across conversations without any changes to your prompts.

​
How wiring works

Engram ships a small MCP server called engram-mcp. It speaks the standard MCP stdio protocol, so any MCP-capable agent can connect to it. When the agent starts, it spawns engram-mcp as a subprocess over stdin/stdout. From that point the agent can:
  • Call the remember tool to stage a new memory for review.
  • Read the memory://recall resource to get a ranked list of relevant memories injected into its context window.
The fastest way to get the correct config snippet for your agent is engram init: 
engram init claude-code   # prints .mcp.json snippet
engram init codex         # prints TOML snippet
engram init opencode      # prints JSON snippet
Run the command, paste the output into the right config file, and restart your agent. The sections below show exactly where each snippet goes.

​
Agent configuration

Place the following in .mcp.json at your project root for a per-project setup, or in ~/.claude.json to wire Engram globally for all Claude Code sessions.
// .mcp.json (project root) or ~/.claude.json
{
  "mcpServers": {
    "engram": { "command": "engram-mcp" }
  }
}
To generate and write this automatically:
engram init claude-code
Claude Code reads .mcp.json on startup and launches engram-mcp as a stdio subprocess. You can verify the connection is live by asking Claude: “What does engram remember about this project?”

​
What the agent can do after wiring

Once engram-mcp is registered, two capabilities become available in every agent session:
CapabilityMCP identifierWhat it does
Store a memoryremember toolStages a candidate fact for human review
Read memoriesmemory://recall resourceReturns ranked, promoted memories for the current project
The agent decides when to call these — you do not need to prompt it explicitly. Most agents will read memory://recall at the start of a conversation and call remember when they learn something worth preserving (a preference, a decision, a constraint).
Memories staged via the remember tool enter a pending queue and are not recalled until you promote them with engram promote. See the Reviewing Memories guide for the full review workflow.

​
Using engram serve for other agents

If you are using an agent or IDE extension that supports MCP but is not listed above, you can point it at engram-mcp directly. The server is a standard stdio MCP process — any client that can launch a local command works. 
engram serve
engram serve starts engram-mcp in the foreground and prints the stdio connection details. Use your agent’s custom MCP configuration UI or config file to register the command engram-mcp (or the full path if it is not on your PATH).
engram-mcp is designed to be spawned on demand by the agent over stdio — you do not need to keep a long-running daemon. Each agent session starts and stops its own subprocess, so there is nothing extra to manage.

​
Next steps

Build docs developers (and LLMs) love

Get started for freeTalk to us