Friday MCP
A self-hosted persistent memory layer for coding agents: decisions, schemas and constraints persist across sessions through MCP.
- 67 ★ on the repository
- Last commit
Extended access
Why this level
- The service stores project decisions, schemas and constraints and exposes an API behind a key.
- Installation runs a network script and binds local ports.
Install
- Works in
Manual install
claude mcp add friday -e FRIDAY_URL="http://localhost:8000" -e BRAIN_API_KEY="your_secret_key" -- python -m mcp.serverStart the Friday service first. Substitute your own values.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The service stores project decisions, schemas and constraints and exposes an API behind a key.
- Installation runs a network script and binds local ports.
This is third-party code. Review the repository files before installing.
What it does
Friday runs as a background service and keeps architecture decisions, schemas and project constraints in four memory tiers. Agents reach it through seven MCP tools: `add_memory` stores decisions with rationale, `add_fact` records versioned facts, `memory_search` searches by meaning, `get_context` compiles a multi-layer context, `get_blast_radius` traverses the dependency graph, and `delete_memory` and `revoke_fact` remove obsolete and conflicting facts. The service starts through Docker Compose: the install script checks Docker, allocates ports 8000, 7474 and 7687, and generates secrets and `.env`. The same server connects to Cursor, Claude Code, Antigravity and VS Code, so memory is shared across tools. A Python SDK and a Neural Studio interface for browsing the knowledge graph are included.
Who it is for. Developers using several agents on one project who lose context in every new session.
Good fit when
- Architecture and schema decisions stay available after the chat closes.
- Before a refactor you want the dependency graph to show which parts it touches.
- Cursor and Claude Code work from one shared project memory.
- Obsolete or conflicting facts get revoked and versioned.
Not a fit when
- Docker is unavailable or you cannot keep a background service and ports 8000, 7474, 7687.
- A rules file in the repository is enough for a small project.
- Memory cannot live on your own server.
Example request
Store in Friday the decision to use PostgreSQL for orders and show which services a schema change would touch.MCP
- Transport
- stdio, http
- Authentication
- API key
Environment variables
FRIDAY_URLrequired- Friday service address, for example http://localhost:8000.
BRAIN_API_KEYrequired, secret- Friday service API secret key.
Limitations
Docker and free ports 8000, 7474 and 7687 are required, and the service keeps project data on your machine or server. Installation runs a script through `curl | bash`: read `install.sh` first. The API is protected by the `BRAIN_API_KEY` secret, so keep it out of the repository. For a remote server the client connects over SSH, and memory accumulates everything the agent stores, including sensitive decisions. Answer quality depends on what the agent stored in memory, so review entries and revoke stale ones with `revoke_fact`.
How to disable. Remove the `friday` entry from the client MCP configuration, stop the service with `docker compose down` in the project folder and delete data volumes if the memory is no longer needed.
FAQ
How do I connect Friday to Claude Code?
Run `claude mcp add friday -e FRIDAY_URL="http://localhost:8000" -e BRAIN_API_KEY="..." -- python -m mcp.server` with your own values.
Which MCP tools are provided?
Seven: `add_memory`, `add_fact`, `memory_search`, `get_context`, `get_blast_radius`, `delete_memory` and `revoke_fact`.
Can I use a remote server?
Yes, the client calls the server over SSH, for example through `docker exec -i fridays-brain-app python /app/mcp_server/server.py`.
Which ports does the service use?
Ports 8000, 7474 and 7687. The install script allocates them itself.
Is there an SDK?
Yes, the `friday-memory` Python package on PyPI. It resolves the address and key from the `FRIDAY_URL` and `FRIDAY_API_KEY` environment.
Discussion
No questions or reviews yet.
Related
MCP serversSelf-hosted long-term memory for agents: semantic search, a knowledge graph and autonomous consolidation over MCP and a REST API
A self-hosted MCP memory server that keeps preferences and decisions across conversations in Claude Code, Cursor and other clients
Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
A self-hosted MCP server for long-term agent memory: structured records, versioned rollback and a visual dashboard
A persistent memory MCP server for agents: observations, beliefs, goals and dream consolidation across 60 tools.
An MCP server and API of shared memory for several agents: facts, decisions and events live in one database with semantic search.
README in short
The repository covers the session amnesia problem, a multi-layer memory architecture, a memory lifecycle and client setup. DeepEval benchmarks and rules for multiple agents are included.