Engram
Give Claude Code and other coding agents ranked local code-graph context on every read and grep without sending the whole repo to the model.
Overview
Engram is an MCP server for the Build phase that ranks local code-graph context and intercepts agent reads and greps.
What is this MCP server?
- Builds a ranked local code graph and injects context when agents read files or run greps
- stdio MCP via npm package engramx (v4.3.2) with Node runtime hint engram-serve
- Optional ENGRAM_MISTAKE_GUARD pre-mortem mode: warn (1) or deny (2) on risky patterns
- HTTP bearer token optional; auto-generated at ~/.engram/http-server.token if unset
- Hooks for Claude Auto-Memory (ENGRAM_ANTHROPIC_MEMORY_PATH) and plugin MCP config (~/.engram/mcp-providers.json)
- Server schema version 4.3.2
- Transport: stdio over npm identifier engramx
- License: Apache-2.0
Community signal: 132 GitHub stars.
What problem does it solve?
Agents waste context windows re-reading whole files and still miss the symbols and call paths that actually matter in your repo.
Who is it for?
Solo builders running Claude Code or Cursor on a medium-to-large local repo who want local, ranked context without a cloud code index.
Skip if: Teams that only need occasional single-file edits or refuse to run a local Node MCP process beside the agent.
What do I get? / Deliverables
After you register Engram, file reads and greps return graph-ranked local context so agents navigate the codebase with less noise and fewer wrong edits.
- stdio MCP server wired into agent read/grep flows
- Local ranked code-graph context on intercepted operations
- Optional pre-mortem guardrails via ENGRAM_MISTAKE_GUARD
Recommended MCP Servers
Journey fit
Engram sits in Build because it wires into the agent’s file-access loop while you are actively implementing and navigating a codebase. Agent-tooling is the canonical shelf: it is an MCP server that reshapes how agents discover and rank symbols, not a deploy or marketing tool.
How it compares
Local code-graph MCP integration, not a generic documentation skill or hosted semantic search product.
Common Questions / FAQ
Who is Engram for?
Indie and solo developers using AI coding agents who want Apache-2.0 local code-graph context on reads and greps.
When should I use Engram?
Use it during active Build work when agents repeatedly miss cross-file relationships or burn tokens scanning entire files.
How do I add Engram to my agent?
Install the npm MCP package engramx, run the engram-serve binary over stdio in your agent’s MCP config, and set ENGRAM_API_TOKEN only if you use the HTTP server.