Cortex Hub MCP
Your own server gives every agent shared memory, AST-level code search and pre-commit checks through one MCP endpoint.
- 106 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The installer runs through curl and bash and installs hooks that affect commits and sessions
- Memory, knowledge and code snippets are stored on a server you administer
Install
- Works in
Manual install
curl -fsSL "https://raw.githubusercontent.com/lktiep/cortex-hub/master/scripts/install.sh" | bashThe command from the One-Command Project Setup section. The script configures MCP and installs hooks, so read it before running. In Claude Code `/install` does the same.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The installer runs through curl and bash and installs hooks that affect commits and sessions
- Memory, knowledge and code snippets are stored on a server you administer
This is third-party code. Review the repository files before installing.
What it does
Cortex Hub builds on your own server what individual agents lack: memory across sessions, shared team knowledge, code search across several projects and quality control. All of it is available through one MCP endpoint that Claude Code, Cursor, Gemini, Codex and others connect to. There are 32 tools in total, including `cortex_session_start` and `cortex_session_end` for sessions, `cortex_code_search`, `cortex_code_context` and `cortex_code_impact` for code, `cortex_memory_store` and `cortex_memory_search` for memory, `cortex_quality_report` and `cortex_plan_quality` for checks. The installer detects IDEs, configures MCP for each, installs hooks into `.claude/hooks/`, creates a project profile and appends to `.gitignore`; it is safe to re-run. The `/install` command does the same inside Claude Code. The server side runs through Docker Compose, and a Conductor distributes tasks between agents with twelve additional tools.
Who it is for. Teams where several people work with different agents and want shared decision memory, code search and unified pre-commit checks.
Good fit when
- After switching laptops the agent should continue where it stopped.
- Knowledge of team mistakes and decisions sits in sessions rather than a shared base.
- Symbol and call search across several projects at once is needed.
- Before a commit the agent must report build, linter and test results.
- Tasks are distributed across several agents through the Conductor.
Not a fit when
- Local memory of one agent without a server is enough.
- You have no Docker and no wish to maintain your own service.
- Running install scripts through `curl | bash` is forbidden by team policy.
Example request
Start a Cortex session, find JWT middleware across all projects and show what would change if the token check function is edited.MCP
- Transport
- http
- Authentication
- API key
Environment variables
CORTEX_API_KEYrequired, secret- Worker agent API key for connecting to your Cortex server.
Limitations
The server must be deployed and maintained by you: Node.js 22 or newer, pnpm 9, Docker 24 and API keys in `infra/.env` are required. Installers run through `curl ... | bash` and set up hooks that interfere with commits and sessions, so read the script before running it. Team memory and knowledge live on your server, so access to it decides who sees decisions and notes. Comparison figures in the README come from the authors. The repository was last updated on 30 October 2026 and uses the MIT licence.
How to disable. Remove the Cortex entry from client MCP settings and the hooks from `.claude/hooks/`, then stop the server stack with `docker compose down` in the `infra` directory. Memory data lives on the server, so delete it separately.
FAQ
How many tools does the server have?
Thirty-two: twenty core ones (sessions, code, memory, knowledge, quality, health) and twelve Conductor tools for task distribution.
What does the installer do?
Detects IDEs (Claude, Gemini, Cursor, Windsurf, VS Code, Codex), configures MCP, installs hooks, creates a project profile and edits `.gitignore`.
How does code search work?
Through `cortex_code_search` over symbols and AST across all projects without naming a repository; `cortex_code_impact` estimates the blast radius.
Is your own server needed?
Yes. The server side starts through Docker Compose from the `infra` directory after copying `.env.example` and adding API keys.
What is the Conductor?
A mode where several agents receive tasks through the `cortex_task_*` and `cortex_agents` tools; a worker agent starts with a separate script.
Discussion
No questions or reviews yet.
Related
MCP serversCortex keeps decisions and project context between Claude Code and Codex sessions in a local database and shows what it recalled
Shared memory for coding agents: messages in SQLite, old context compacted into a summary tree and MCP tools to search it.
A memory and coordination layer for coding agents: specs and issues live in git while agents work in parallel
A persistent memory MCP server for agents: observations, beliefs, goals and dream consolidation across 60 tools.
MCP server of the self-hosted EmperorClaw panel: Claude Code and Codex read company knowledge, tasks and customers in scope
Local memory for agents: stores conversations and project files verbatim and retrieves them with semantic search over MCP
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Cortex Hub (this card) | no | API key | 106 | |
| agentmemory | yes | not required | 30,314 | local memory for developer agents based on hooks; here server-side shared memory with code search and quality checks |
| codebase-memory-mcp | no | not required | 48,531 | an MCP server that builds a code graph and answers locally; here the code graph is combined with memory, knowledge and commit control |
README in short
The README explains the problem of isolated agents, shows a diagram with one MCP endpoint, a table of tools and three quick-start scenarios.