edkarlsson avatar

Locus

edkarlsson/locus

Gives Claude a hierarchical markdown file system designed around the method of loci: each directory is a "room" containing specific knowledge, and agents read only what they need instead of loading everything. Exposes five MCP tools over stdio: memory_list (navigate rooms), memory_read, memory_write (session logs only, canonical files are protected), memory_search (ripgrep or fallback), and memory_batch (read up to 20 files at once). Includes optional Ed25519 signing to tag outputs as trusted or data before the agent sees them. Built for keeping context windows small when working across multiple projects or knowledge domains. Benchmarks show 52% fewer context lines loaded for specific queries compared to flat memory files.

tools
5
transport
STDIO
last updated
Mar 3, 2026

Tools

5 tools

Public tool metadata for what this MCP can expose to an agent.

memory_listReturns INDEX.md (no args) or lists a room's files

Returns INDEX.md (no args) or lists a room's files

No parameter schema in public metadata yet.

memory_readReads any file in the palace

Reads any file in the palace

No parameter schema in public metadata yet.

memory_writeAtomically writes a file (guarded — cannot write to _metrics/, sessions/, .sig/, .security/)

Atomically writes a file (guarded — cannot write to _metrics/, sessions/, .sig/, .security/)

No parameter schema in public metadata yet.

memory_searchRanked full-text search over the shared FTS5 index (see Recall); ripgrep only without FTS5

Ranked full-text search over the shared FTS5 index (see Recall); ripgrep only without FTS5

No parameter schema in public metadata yet.

memory_batchReads up to 20 palace files in a single call — use for multi-room loads

Reads up to 20 palace files in a single call — use for multi-room loads

No parameter schema in public metadata yet.

Sponsors

Readme

Locus

CI PyPI version License: MIT Python 3.11+

Hierarchical markdown-based memory system for autonomous AI agents. Each directory is a room (locus) in the palace, containing specific knowledge navigated on demand. Named for the atomic unit of the Method of Loci.

Core idea: Keep context windows small. Load only the room you need, not the whole palace.


How it works


View the full README on GitHub

Details

Repository
github.com/edkarlsson/locus
Categories
AI & LLM Tools, Documents & Knowledge
Package
locus-mcp
Transport
STDIO
Registry
MCP registry, active

Related AI & LLM Tools MCP servers

View all
1egoughnour avatarMassive Context Mcpegoughnour/massive-context-mcpHandles 10M+ token contexts with chunking, sub-queries, and local Ollama inference.0stars
2eltociear avatarSecrets Audit Mcpeltociear/secrets-audit-mcpDetects leaked secrets & API keys: 32+ provider rules (AWS, GitHub, Stripe, OpenAI…), zero deps.0stars
3freeguy-ai avatarCloseread Verifyfreeguy-ai/closeread-verifyA verified dependency-audit verdict any AI agent can call over MCP, not a raw scan.0stars
4freminder avatarESG MCP Serversfreminder/esg-mcp-servers31 MCP tools for ESG data extraction, PDF processing, vector search, and EU regulation analysis.0stars
5getcyphrex avatarCyphrexgetcyphrex/platformChecks AI agent actions before they run. Signed, RFC 3161 timestamped evidence for audits.0stars
6goatgaucho avatarTrustLayergoatgaucho/trust-layerAgent reputation scoring: trust scores, Sybil detection, cross-chain identity for 133K+ agents0stars