Locus
Give Claude Code or Cursor a hierarchical markdown memory palace so long-running agent work stays navigable instead of one flat context dump.
Overview
Locus is an MCP server for the Build phase that gives AI agents hierarchical markdown memory-palace navigation through stdio MCP tools.
What is this MCP server?
- Hierarchical markdown memory palace model for agent recall
- MCP tools for structured palace navigation and reads
- stdio transport via PyPI package locus-mcp (v0.6.2)
- GitHub source at EDKarlsson/locus for self-hosted or fork workflows
- Server version 0.6.2 on PyPI identifier locus-mcp
- Transport: stdio
- Repository: github.com/EDKarlsson/locus
What problem does it solve?
Agents lose thread on large codebases because context is flat and ephemeral, so builders repeat explanations and lose decision history.
Who is it for?
Solo builders running Claude Code or Cursor on multi-session agent workflows who want local, markdown-native structured memory.
Skip if: Teams that need a managed cloud memory SaaS, full-text enterprise search, or zero local Python/MCP setup.
What do I get? / Deliverables
After registering Locus, your agent can navigate a structured markdown palace via MCP so long tasks retain locatable memory without stuffing the whole repo into every prompt.
- stdio MCP server exposing palace navigation tools to your agent
- Structured markdown memory graph the agent can query across sessions
- Local, self-hosted memory layer without a third-party memory API
Recommended MCP Servers
Journey fit
How it compares
MCP memory-palace integration, not a Claude skill or hosted RAG platform.
Common Questions / FAQ
Who is Locus for?
Indie and solo builders using MCP-enabled coding agents who want hierarchical markdown memory they control locally.
When should I use Locus?
Use it while building agent-driven features or long refactors when flat chat context and scattered notes are breaking continuity.
How do I add Locus to my agent?
Install the locus-mcp PyPI package (0.6.2), configure stdio MCP in Claude Code or your client, and point tools at your palace markdown layout per the GitHub repo.