Engraphis MCP
Engraphis keeps a coding agent's memory in a local SQLite database and returns an answer with its source, or abstains honestly.
- 183 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Stores facts about project code and decisions in a local SQLite database reachable by any process with user rights
- With allow_remote=true and a connected cloud account, some queries and decisions leave the local machine
Install
- Works in
Manual install
python -m pip install "engraphis[all]"Installs the local app with the engraphis-dashboard panel; needs Python 3.10 or newer.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Stores facts about project code and decisions in a local SQLite database reachable by any process with user rights
- With allow_remote=true and a connected cloud account, some queries and decisions leave the local machine
This is third-party code. Review the repository files before installing.
What it does
Engraphis installs as a Python package and keeps project memory in a SQLite database on the user's disk, with no cloud account needed for local use. engraphis-dashboard opens a web panel at 127.0.0.1:8700 that shows the recorded facts and the graph of relations between them. An agent saves memory with engraphis_remember and retrieves it with engraphis_recall_context, scoped by workspace, repository and session. Memory search combines vector, lexical, graph and code-aware retrieval, packing the result into a fixed token budget. When recalling a fact, Engraphis either cites the memories that support it or explicitly abstains if the evidence is too weak. History is bi-temporal: an old version of a fact is not erased but closed, and it can still be pulled up with a separate timeline query. An agent connects over MCP: the local server starts with engraphis-mcp from the engraphis[mcp] package and runs over stdio, or as the separate engraphis-mcp-http process on the loopback address 127.0.0.1:8711 for clients that need HTTP. By default the agent gets nine compact Smart MCP tools and discovers less common operations itself, with no manual profile choice.
Who it is for. Developers who run several projects with the same coding agent and want it to remember why past decisions were made across sessions without manual notes.
Good fit when
- The agent keeps re-deriving why auth moved from JWT to PASETO instead of recalling it from a past session
- Session context needs restoring at the start of the day without re-explaining the project to the agent
- It matters when and why a stored fact changed, not just its current value
- Project code needs searching and linking to memory by meaning, not only by keyword
- Several workspaces on one machine need separately scoped memory, without mixing repositories
Not a fit when
- Several people or remote agents need one shared hosted team memory: shared Team memory and hosted MCP are not available yet
- Memory needs to sync across machines without a paid plan: Cloud Sync is a Pro and Team feature
- A memory engine is needed on a machine without Python 3.9 or newer
- The goal is a ready answer for the user directly: Engraphis returns context and sources to the agent, not a final text
Example request
Connect Engraphis to this repository over MCP and remember that auth moved from JWT to PASETO this sprintMCP
- Transport
- stdio, http
- Authentication
- key is optional
Environment variables
ENGRAPHIS_API_TOKENsecret- Token protecting the local memory HTTP server; needed only if it is exposed beyond loopback.
Limitations
- The full local engine with the dashboard needs Python 3.10 or newer; the NumPy-only core works from Python 3.9.
- Shared Team memory and hosted MCP for a team are marked not yet available by the developer, though Team roles and a 30-day account activity log already exist.
- The paid Jev assistant adds up to 500 calls per person per day on Pro and Team, advising on risky commands and contradictory memories, but it cannot write memories itself, change grounding requirements, or certify that a task is complete.
- Auto Consolidation and Auto Dreaming on hosted plans produce proposals rather than applying changes themselves; local governance still decides.
How to disable. Remove the MCP server from the agent's config and stop the engraphis-mcp-http background process if it was running. Uninstall with pip uninstall engraphis and delete the engraphis.db file to remove the memory entirely.
FAQ
Does Engraphis need a cloud account to work?
No, one person on one machine only needs pip install "engraphis[mcp]" and the local MCP server. A cloud account matters only for Pro and Team, when sync or team sharing is needed.
How is Engraphis different from an agent's plain notes file?
A notes file gets read whole and is not searched by meaning. Engraphis combines vector, lexical, graph and code-aware search, fits the answer into a set token budget, and either shows the fact's source or abstains when the evidence is weak.
What happens to an old version of a fact after an edit?
It is not deleted but closed, marked with when it stopped being true. A timeline query with closed_only=true pulls up only those closed versions, and current memory for other queries stays untouched.
How do I connect Engraphis to an MCP client that needs HTTP instead of stdio?
Run engraphis-mcp-http from the same engraphis[mcp] package: the server listens only on the loopback address 127.0.0.1:8711 and skips the dashboard's auth layer, so it should never be exposed to the network directly.
What does the paid Jev assistant add?
On Pro and Team it reviews up to 500 calls per person per day for free: judging command risk, finding contradictory memories and checking answer grounding, while the decision and the memory write stay with the user and the local engine.
Discussion
No questions or reviews yet.
Related
MCP serversEngram by Gentleman Programming
Engram (Gentleman Programming)
Persistent memory for AI coding agents in a single Go binary: an MCP server on SQLite with full-text search that remembers your project across sessions
Engram (getengram.app): cloud memory for agents
Engram (getengram.app)
A remote MCP server that keeps agent conversations verbatim and searches them by meaning across sessions
Local project memory for agents: constraints, decisions and lessons live in an evidence-backed tree and return before a task.
Self-hosted long-term memory for agents: semantic search, a knowledge graph and autonomous consolidation over MCP and a REST API
Piia Engram: portable memory for AI agents
Piia Engram
Local memory about you and your projects: Claude Code, Codex and Cursor read the same preferences and lessons from your disk
A local-first MCP memory server for agents: session history, codebase index and a concept graph in one SQLite database
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Engraphis (this card) | yes | key is optional | 183 | |
| MCP Memory Service | no | not required | 2,117 | bi-temporal fact history with abstention on weak evidence, not just note storage and search |
| Basic Memory | no | not required | 4,566 | combines vector, lexical, graph and code-aware search within one token budget |
README in short
The front page frames memory as continuity across sessions, useful retrieval from four search sources, grounded answers with history, and a Spec Studio for checking prompt specs against scoped memory. It separately lists three offline fixtures: context reduction from structure-aware chunking, a smaller JSON payload estimate and the share of grounded versus abstained answers, with a link to the benchmark methodology and a note that these are fixed samples, not a general task-performance claim.
README badge
Original SKILL.md
--- name: engraphis-memory description: 'Give the agent durable, scoped, explainable memory across sessions and repositories through the Engraphis MCP tools. Use when you learn a convention, decision, bug cause/fix, or user preference worth keeping; when prior context would help before you answer or act (to avoid re-asking or re-deriving); when asked "why is it like this" or "how has this changed over time"; or when starting or resuming work in a repo. Triggers: remember, recall, "what do we know about X", why/rationale, timeline/history, retire/pin/correct, session handoff, index/search code.' --- # Engraphis Memory Engraphis is a local-first memory engine exposed to agents over MCP. This skill is the *discipline* for using it well: what to store, how to scope it, and which tool answers which question.