MCP integration

Your coding agent forgets your codebase every time you close the tab.

You spent 20 minutes explaining your project to Claude. Tomorrow it won't remember any of it. The Hebbrix MCP server fixes that for good. Set it up once, and your agent keeps your architecture decisions, conventions, bug fixes, and project context across every session.

Works with Claude Desktop, Cline, Cursor, and any MCP-compatible agent.

Setup guide →Get your API key

one-time setup

Three minutes, then you're done

Install the server, add two lines to your Claude Desktop config, and restart the app. That's the whole setup, and you'll never re-explain your project again.

Install the MCP server

Clone the repo and run the quick setup script. Python 3.9+ required, no other dependencies.

Add to your agent config

Two lines in claude_desktop_config.json: the command path and your Hebbrix API key. Save and restart.

Start teaching your agent

Tell it something worth remembering. It stores it instantly, and it's still there next session, next week, and next year.

claude_desktop_config.json

{
  "mcpServers": {
    "hebbrix": {
      "command": "python",
      "args": [
        "/path/to/hebbrix/mcp/server.py"
      ],
      "env": {
        "HEBBRIX_API_KEY": "your_api_key"
      }
    }
  }
}

# Restart Claude Desktop. Done.
# Your coding agent now has
# persistent memory.

four tools

What your agent can do now

Four MCP tools. Each one gives your agent a capability it didn't have before.

hebbrix_remember

Remember anything from the conversation

"Remember that we use PostgreSQL with Prisma ORM, and all API routes go through the auth middleware." Your agent stores this, and it's still there next week and next month.

hebbrix_search

Search across everything it has learned

"What database pattern do we use?" 5-layer hybrid search finds the right memory even if you phrase it differently than when you stored it.

hebbrix_recall

Get context for the current task

Surfaces the relevant project context for whatever you're working on. You don't have to prompt for it. The agent knows what to pull.

hebbrix_forget

Remove outdated information

"Forget that we use REST, we migrated to GraphQL last sprint." Clear out stale context so your agent stops giving outdated advice.

a week with memory

What it actually feels like

When your coding agent remembers your project, conversations stop repeating themselves.

A Cursor chat where the agent uses a Hebbrix MCP tool to recall a stored architecture decision

Monday

You explain your auth architecture

JWT tokens, refresh rotation, middleware validation. The agent stores the whole pattern, not a summary, including the specific decisions and why you made them.

Wednesday

New protected endpoint, no explanation needed

You ask Claude to add a new route. Without you saying anything, it uses your JWT pattern, adds the middleware, and follows your error handling style, because it already knows it.

Friday

A bug. Claude already knows where to look.

Auth issue. Claude remembers Monday's implementation, including the quirk where the token refresh overlaps with rate limiting. It finds the bug in seconds instead of hours.

Next week

New sprint, same context

New conversation, same agent brain. It knows your bug fix, your auth pattern, your team's coding style, and your deployment process. It picks up right where you left off.

Give your coding agent a memory it keeps

One config file. Five minutes. Never explain your project twice.

Setup guide →Get API key