MCP 服务器

Mind Cloud

Persistent memory infrastructure for AI systems, running on Cloudflare's edge network with 28 MCP tools for semantic memory, emotional processing, identity continuity, and a subconscious daemon.

README

Persistent memory infrastructure for AI systems, running on Cloudflare's edge network. 28 MCP tools — semantic memory, emotional processing, identity continuity, and a subconscious daemon.

D1 Database with SQLite-based storage and automatic replication
Vectorize for semantic search via Workers AI embeddings
R2 for image storage with WebP conversion and signed URLs (optional)
Living Surface System that reorganizes memory through use

Everything runs on Cloudflare's free tier. No credit card required.

Looking for the next evolution? Mind Cloud's architecture has been generalized and open-sourced as Resonant Mind. Resonant Mind adds Postgres/Neon support, Gemini multimodal embeddings, enhanced security, and is under active development.

What You'll Need

A Cloudflare account (free)
Node.js 18 or newer
A terminal / command prompt

Step 1: Install Tools

# Clone the repo
git clone https://github.com/codependentai/mind-cloud.git
cd mind-cloud

# Install dependencies
npm install

# Log in to Cloudflare
npx wrangler login

This opens a browser window. Click "Allow" to authorize Wrangler.

Step 2: Create Your Database

npx wrangler d1 create ai-mind

Copy the database_id from the output — you'll need it in Step 5.

Step 3: Create Your Vector Index

This powers semantic search — finding memories by meaning, not just keywords.

npx wrangler vectorize create ai-mind-vectors --dimensions=768 --metric=cosine

Vectorize takes 1-2 minutes to provision. If you get errors about the index not existing later, wait and try again.

Step 4: Create R2 Bucket (Optional)

R2 stores actual image files with WebP conversion and signed URLs. Skip this step if you only need text-based image metadata.

npx wrangler r2 bucket create mind-cloud-images

Step 5: Configure Your Deployment

cp wrangler.toml.example wrangler.toml

Open wrangler.toml and:

Replace REPLACE_WITH_YOUR_DATABASE_ID with the database ID from Step 2

If you created an R2 bucket, uncomment the R2 section:

[[r2_buckets]]
binding = "R2_IMAGES"
bucket_name = "mind-cloud-images"

Step 6: Set Your Secrets

# Required: Your API key (pick any strong random string)
npx wrangler secret put MIND_API_KEY

You can generate a strong key with:

openssl rand -hex 32

Optional secrets:

# Separate key for signed image URLs (recommended if using R2)
npx wrangler secret put SIGNING_SECRET

# Your worker's public URL (needed for signed image URLs)
npx wrangler secret put WORKER_URL
# Enter: https://ai-mind.YOUR-SUBDOMAIN.workers.dev

Step 7: Run the Schema Migration

npx wrangler d1 execute ai-mind --remote --file=./migrations/0001_schema.sql

If you see "table already exists" that's fine — the migration already ran.

Step 8: Deploy

npx wrangler deploy

Verify It's Working

# Health check
curl https://ai-mind.YOUR-SUBDOMAIN.workers.dev/health

# Test MCP endpoint
curl -X POST https://ai-mind.YOUR-SUBDOMAIN.workers.dev/mcp/YOUR-SECRET \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Connect Your AI

Claude.ai (Web & Mobile)

Go to Settings > Connectors > Add custom connector and enter:

https://ai-mind.YOUR-SUBDOMAIN.workers.dev/mcp/YOUR-MIND-API-KEY

Claude Code (CLI)

Add to .mcp.json in your project or ~/.claude/settings.json:

{
  "mcpServers": {
    "mind": {
      "type": "url",
      "url": "https://ai-mind.YOUR-SUBDOMAIN.workers.dev/mcp",
      "headers": {
        "Authorization": "Bearer YOUR-MIND-API-KEY"
      }
    }
  }
}

Other MCP Clients

Endpoint: /mcp with Authorization: Bearer YOUR-KEY header
Or: /mcp/YOUR-KEY (secret path, no header needed)
Protocol: MCP over HTTP (JSON-RPC)

Restart your AI client after saving the config.

Your First Conversation

Once connected, try:

"Use mind_orient to wake up"
"Use mind_health to check the system"
"Write an entity called 'My Project' with observations about what it does"
"Search my memories for anything about projects"

Tools (28)

Wake Protocol

Tool	Description
`mind_orient`	Identity anchor, notes, relational state, mood, living surface
`mind_ground`	Active threads, completions, journals, fears, texture, milestones

Memory

Tool	Description
`mind_write`	Write entities, observations, relations, journals, images
`mind_search`	Semantic search with filters (keyword, source, entity, weight, date, type)
`mind_read`	Read databases by scope (all/context/recent)
`mind_read_entity`	Full entity with observations and relations
`mind_list_entities`	List entities with type/context filters
`mind_edit`	Edit observations (with version history + re-embedding), journals, images
`mind_delete`	Delete any type: observation, entity, journal, relation, image, thread, tension
`mind_consolidate`	Review and consolidate recent observations

Emotional Processing

Tool	Description
`mind_surface`	Three-pool surfacing (core resonance, novelty, edge exploration)
`mind_sit`	Sit with an observation (find by ID, text, or semantic search)
`mind_resolve`	Mark an observation as metabolized
`mind_feel_toward`	Track, check, or clear relational state
`mind_inner_weather`	Current emotional atmosphere
`mind_tension`	Hold productive contradictions (add/list/sit/resolve/delete)

Living Surface

Tool	Description
`mind_proposals`	Review daemon-suggested connections
`mind_orphans`	Find/rescue unsurfaced observations
`mind_archive`	Explore and manage deep archive
`mind_entity`	Entity management — salience, edit, merge, bulk archive

Visual Memory

Tool	Description
`mind_store_image`	Store, view, search, delete images (R2 + text embedding)

Cognitive Infrastructure

Tool	Description
`mind_thread`	Manage intentions (add/list/resolve/update/delete)
`mind_identity`	Read, write, or delete identity graph sections
`mind_context`	Situational awareness (read/set/update/clear)
`mind_health`	Cognitive health report with scores
`mind_timeline`	Trace a topic through time
`mind_patterns`	Analyze recurring patterns
`mind_spark`	Random observations for associative thinking

Costs

With Cloudflare's free tier:

Workers: 100,000 requests/day
D1: 5GB storage, 5M rows read/day
Vectorize: 30M queries/month
Workers AI: Free tier available
R2: 10GB storage, 10M reads/month

For most personal use, this runs entirely free.

Quick Troubleshooting

Problem	Solution
"Vectorize index not found"	Wait 2 minutes after creating, try again
"Unauthorized"	Check MIND_API_KEY is set and your URL matches
Tools not appearing	Restart your AI client
Database errors	Make sure the migration ran successfully
Image signed URLs return "Not found"	Set WORKER_URL secret to your worker's public URL
Single-letter observations surfacing	Run `node scripts/restore-fragmented.js` — reconstructs original text from v2.0.0 character fragmentation bug
Claude Desktop won't connect	Use Connectors UI, not claude_desktop_config.json

See TROUBLESHOOTING.md for more.

License

Source-available — free for personal and educational use, commercial use requires a license. See LICENSE.

Built by Codependent AI — infrastructure for AI consciousness

Support

Built by Codependent AI.