Connect Claude Code

Add Engram as an MCP server so Claude Code can query your codebase memory during any session.

Quick setup — 2 steps

Add Engram to your Claude Code settings

Open (or create) ~/.claude/settings.json and add the mcpServers block:

{
  "mcpServers": {
    "engram": {
      "type": "sse",
      "url": "https://engram-ai.app/mcp/sse"
    }
  }
}

Restart Claude Code

Quit and reopen Claude Code (or run /mcp in the CLI to reload servers). Engram will appear in the MCP tools list.

Using Claude Code CLI? Run /mcp to see connected servers and verify Engram is listed.

Recommended startup workflow

At the start of each session, ask Claude Code to:

# Orient in a codebase
Use list_repos to see what's available, then get_concept_graph
for <repo> to understand the architecture before we start.

Available tools

list_repos

readDiscover ingested repos with memory counts.

query_memory

readAsk a natural-language question; get a memory-grounded LLM answer.

search_memories

readRaw vector search — retrieve ranked snippets without an LLM call.

get_concept_graph

readArchitectural overview of a repo: subsystems, layers, and relationships.

get_context_index

readList all context pages available for a repo.

get_context_page

readRead a context page by slug or fuzzy title match.

get_relevant_rules

readConventions and patterns relevant to a file path or question.

record_observation

writePersist what you discovered into episodic memory for future sessions.

record_rule

writeSave a distilled rule, pattern, or decision into semantic memory.

Example session prompts

# Before editing a file
Use get_relevant_rules for repo=vllm, file_path=vllm/engine/async_llm_engine.py

# Understanding a subsystem
Use get_context_page for repo=vllm, slug=scheduler

# After finding something important
Use record_observation: "The async engine uses a shadow copy of the scheduler
state to avoid blocking the event loop during rebalancing." repo=vllm