MCP-Compatible · Open Source · Apache 2.0

Claude.md is for context.
MarsNMe is for continuity.

Your AI tools should know you — not start from scratch every time. MarsNMe gives your AI persistent memory across sessions and tools, so you and your AI grow together.

13MCP Tools
Profiles
1024Embedding Dims
Memory Horizon
MarsNMe demo — AI memory recall in action

Context is what you write.
Continuity is what you live.

Claude.md records what you remember to write. MarsNMe records what actually happened — the decisions, the doubts, the near-misses. That's not context injection. That's a relationship.

Your AI finally remembers you — MarsNMe
Without vs With MarsNMe
How MarsNMe works

Two layers of memory,
one coherent mind.

01

Short-term Memory

Session context and recent interactions stored in <profile>.memories. Fast reads, ~30 day TTL.

02

Long-term Knowledge

Semantic chunks ingested into marsvault_chunks with Jina v3 embeddings. Semantic recall via pgvector.

03

Multi-Profile Architecture

One codebase, any number of agents. Set MCP_PROFILE=my-agent to route all reads and writes to isolated schemas.

04

Schema-as-Code

All migrations live in supabase/migrations/. Reproducible deployments, version-controlled memory schema.

05

Cloud Hosted or Self-Hosted

Use mcp.marsnme.com/your-name for zero-config cloud routing, or deploy marsnme-local to your own Cloudflare account for full self-hosting with D1 + Vectorize.

MarsNMe — multi profile
# Start your agent profile $ MCP_PROFILE=my-agent npx @marsnme/mcp-gateway ✓ MarsNMe MCP Gateway listening :18790   # Or run a second agent on another port $ MCP_PROFILE=assistant-b PORT=18791 npx @marsnme/mcp-gateway ✓ MarsNMe MCP Gateway listening :18791   # Health check $ curl http://localhost:18790/health {   "status": "ok",   "profile": "my-agent",   "tools": 13,   "uptime": 42 }   $

Everything an agent needs
to remember.

Semantic Recall

Jina embeddings v3 with 1024 dimensions. Find memories by meaning, not just keywords. pgvector HNSW index for millisecond retrieval.

jina-embeddings-v3
Multi-Profile Isolation

Each agent gets its own schema. Same codebase, zero cross-contamination. Run unlimited profiles with isolated memory spaces.

MCP_PROFILE
Two-Tier Memory

Short-term session memory with TTL and long-term knowledge chunks that persist forever. Agents know what just happened and what always matters.

memories · marsvault_chunks
MCP Native

Full MCP protocol: initialize, tools/list, tools/call, ping. Drop into any MCP-compatible client — Cursor, Claude Desktop, Warp, or your own agent.

Model Context Protocol
Bearer Auth

Optional Bearer token enforcement via MCP_REQUIRE_BEARER. Secure your endpoint for production use.

production-ready
One Command Install

Install via npm. No cloning, no build steps. npx @marsnme/mcp-gateway and you're running.

npm · npx
Agent-Assisted Install

One prompt.
Your agent installs the rest.

Paste the prompt below into your AI assistant. It will set up Supabase, configure your environment, and start MarsNMe automatically.

① Paste agent prompt

Copy the install prompt and paste into your AI coding assistant. The agent handles Supabase setup, npm install, and environment config.

② Fill your keys

Provide your SUPABASE_BASE_URL, SUPABASE_SERVICE_ROLE_KEY and JINA_API_KEY. Agent writes your .env file.

③ Verify health

Agent runs curl http://127.0.0.1:18790/health to verify the gateway responds correctly.

④ Configure your client

Add the MCP endpoint http://127.0.0.1:18790/mcp to your client config. Start remembering.

Agent Install Prompt
# Paste this into Cursor Agent (Cmd+L → Agent mode)

Please install MarsNMe memory backend for my AI agent:
1. Read the Quick Start at https://github.com/Marsmanleo/MarsNMe#quick-start
2. Create a Supabase project (free plan) at https://supabase.com/dashboard/new
3. Get a Jina API key at https://jina.ai/api-key/
4. Run migrations from supabase/migrations/ in filename order
5. Start: MCP_PROFILE=my-agent npx @marsnme/mcp-gateway
6. Verify: curl http://127.0.0.1:18790/health
7. Add to Cursor MCP settings:
   URL: http://127.0.0.1:18790/mcp
Report health status when done.
Cursor Agent will handle the entire setup interactively.
It will ask for your keys at step 3 — never paste them directly.

13 tools. Everything
your agent needs.

insert_memory

Write short-term memory SHORT

Store session context and recent interactions. Auto-routes to your profile's memories table. ~30 day TTL.

list_memories

List recent memories SHORT

Retrieve recent short-term memories in reverse chronological order. Supports limit and source filters.

search_memories

Semantic search short-term SHORT

Jina-powered semantic search over session memories. Returns similarity scores. Ideal for contextual lookups.

recall

Semantic recall long-term LONG

Deep semantic search over marsvault_chunks. Routes to your profile schema. Supports type, visibility, and similarity filters.

memory_ingest

Ingest long-term insight LONG

Chunk and embed insight content into marsvault_chunks. Supports visibility (private/shared/global), tags, and custom chunk sizes.

dream_ingest

Ingest digest content LONG

Specialized tool for ingesting structured digest reports as long-term knowledge. Auto-chunks and indexes for semantic retrieval.

session_boot

Session startup rhythm OPS

One-call session boot: identity recall, workflow status, health snapshot, and heartbeat sign-in. Run at session start.

session_close

Session close summary OPS

Store session close summary with configurable retention. Captures topics, mood, and key outcomes for next session.

health_check

Memory diagnostics OPS

Run coverage map, expiry alerts, and conflict detection across your memory tables. Essential for maintenance.

reload_source_registry

Reload source registry OPS

Refresh the memory source whitelist from configuration. Add or remove sources without restarting the gateway.

demote_memory

Demote long-term chunk LONG

Reduce the priority or visibility of a long-term memory chunk. Soft downgrade without permanent deletion.

soft_forget

Soft forget memory OPS

Mark a memory as forgotten without deleting it. The memory stays in storage but is excluded from normal recall.

explain_memory

Explain memory content OPS

Get a human-readable explanation of a specific memory entry. Shows metadata, source, tags, and relevance context.

Environment Variables

MCP_PROFILErequired
Your profile identifier. Routes all reads/writes to the corresponding schema. Use any name you want.
SUPABASE_BASE_URLrequired
Supabase REST API project URL (e.g. https://xxx.supabase.co).
SUPABASE_SERVICE_ROLE_KEYrequired
Supabase service role key with schema access. Never expose in client-side code.
JINA_API_KEYrequired
Jina AI API key for generating embeddings. Required for search_memories, recall, and ingest tools.
PORToptional
Default: 18790. Override for running multiple profiles on different ports.
MCP_REQUIRE_BEARERoptional
Set true for production. Enforces Bearer token auth on all MCP endpoints.

Your AI deserves to
remember you.

Apache 2.0. Self-host, fork, extend. Built by Mars Group — a Macau company building the future of human-AI symbiosis.

Join the waitlist

Be the first to know about new features, integrations, and the future of AI memory.

No spam. Unsubscribe anytime. Powered by Listmonk.

MarsNMe — Not Just Context. Continuity. | Product Hunt