Build Your Cerebro

What You're Building

A cloud-hosted database that turns your raw thoughts into searchable, AI-accessible knowledge. You'll set up a PostgreSQL instance with vector embeddings, deploy an MCP server on top of it, and connect your favourite AI tools so they can read and write to a shared memory.

What You Need

Roughly 30 minutes. No programming required — every step is copy-and-paste.

Services (All Free Tier)

Supabase — Hosts the database and runs the server-side functions
OpenRouter — Routes embedding and metadata-extraction calls to the right AI models

Cost

Component	Typical Cost
Supabase (free tier)	$0
Embeddings via text-embedding-3-small	~$0.02 per million tokens
Metadata extraction via gpt-4o-mini	~$0.15 per million input tokens

At around 20 thoughts per day, expect roughly $0.10–0.30 per month in OpenRouter API charges.

Credential Tracker

Throughout this guide you'll create accounts and generate keys. Keep them in one place so you can reference them later. Copy the block below into any text editor:

CEREBRO CREDENTIALS
═══════════════════════════════

SUPABASE
  Email:              ____________
  Database password:  ____________  ← Step 1
  Project ref:        ____________  ← Step 1
  Project URL:        ____________  ← Step 3
  Service role key:   ____________  ← Step 3

OPENROUTER
  API key:            ____________  ← Step 4

CEREBRO
  MCP Access Key:     ____________  ← Step 5
  Server URL:         ____________  ← Step 6
  Connection URL:     ____________  ← Step 6

═══════════════════════════════

Step 1: Create Your Supabase Project

Supabase provides the PostgreSQL database that stores your thoughts and the Edge Function runtime that hosts the MCP server.

Go to supabase.com and create an account (GitHub login works)
In the dashboard, click New Project
Choose an organization (the default is fine)
Name the project cerebro (or your preference)
Generate a database password — copy it to your credential tracker now
Select the region closest to you
Click Create new project and wait for provisioning (~1–2 minutes)

Your project ref is the alphanumeric string in the dashboard URL: supabase.com/dashboard/project/abcxyz123. Copy it to your tracker.

Step 2: Set Up the Database

Enable the Vector Extension

In the left sidebar: Database → Extensions → search for "vector" → flip pgvector ON.

Create the Thoughts Table

In the left sidebar: SQL Editor → New query → paste and Run:

create table thoughts (
  id uuid default gen_random_uuid() primary key,
  content text not null,
  embedding vector(1536),
  metadata jsonb default '{}'::jsonb,
  created_at timestamptz default now(),
  updated_at timestamptz default now()
);

-- Index for fast vector similarity search
create index on thoughts
  using hnsw (embedding vector_cosine_ops);

-- Index for filtering by metadata fields
create index on thoughts using gin (metadata);

-- Index for date range queries
create index on thoughts (created_at desc);

-- Auto-update the updated_at timestamp
create or replace function update_updated_at()
returns trigger as $$
begin
  new.updated_at = now();
  return new;
end;
$$ language plpgsql;

create trigger thoughts_updated_at
  before update on thoughts
  for each row
  execute function update_updated_at();

Create the Search Function

New query → paste and Run:

create or replace function match_thoughts(
  query_embedding vector(1536),
  match_threshold float default 0.7,
  match_count int default 10,
  filter jsonb default '{}'::jsonb
)
returns table (
  id uuid,
  content text,
  metadata jsonb,
  similarity float,
  created_at timestamptz
)
language plpgsql
as $$
begin
  return query
  select
    t.id,
    t.content,
    t.metadata,
    1 - (t.embedding <=> query_embedding) as similarity,
    t.created_at
  from thoughts t
  where 1 - (t.embedding <=> query_embedding) > match_threshold
    and (filter = '{}'::jsonb or t.metadata @> filter)
  order by t.embedding <=> query_embedding
  limit match_count;
end;
$$;

Lock Down Security

One more new query:

alter table thoughts enable row level security;

create policy "Service role full access"
  on thoughts
  for all
  using (auth.role() = 'service_role');

Verify

Open the Table Editor in the sidebar — you should see a thoughts table with six columns (id, content, embedding, metadata, created_at, updated_at). Then check Database → Functions and confirm match_thoughts appears in the list.

Step 3: Save Your Connection Details

Open the Supabase sidebar: Settings (gear icon) → API. Grab these two values and paste them into your credential tracker:

Project URL — shown at the top of the API page
Secret key — listed under "API keys" (previously called "Service role key"). Click the eye icon to reveal it, then copy.

⚠️ The secret key grants full database access. Keep it private — don't share it or commit it to a repo.

Step 4: Get an OpenRouter API Key

OpenRouter acts as a single gateway to many AI models. Cerebro uses it for both embedding generation and metadata extraction.

Sign up at openrouter.ai
Navigate to openrouter.ai/keys
Click Create Key and name it cerebro
Copy the key to your credential tracker right away
Under Credits, add $5 (this covers months of normal usage)

Step 5: Create an Access Key

Your MCP server sits behind a public URL. To prevent unauthorized access, you'll generate a random key that the server checks on every request.

Run one of these commands in your terminal to produce a 64-character hex string:

Mac/Linux:

openssl rand -hex 32

Windows (PowerShell):

-join ((1..32) | ForEach-Object { '{0:x2}' -f (Get-Random -Maximum 256) })

Save the output in your credential tracker under MCP Access Key — you'll set it as a Supabase secret in the next step.

Step 6: Deploy the MCP Server

Install the Supabase CLI

Mac (Homebrew):

brew install supabase/tap/supabase

Windows (Scoop):

scoop bucket add supabase https://github.com/supabase/scoop-bucket.git
scoop install supabase

Linux:

npm install -g supabase

Log In and Link

supabase login
supabase link --project-ref YOUR_PROJECT_REF

Set Your Secrets

supabase secrets set MCP_ACCESS_KEY=your-access-key-from-step-5
supabase secrets set OPENROUTER_API_KEY=your-openrouter-key-here

Create and Deploy the Function

supabase functions new cerebro-mcp

Copy the contents of integrations/mcp-server/deno.json into supabase/functions/cerebro-mcp/deno.json.

Copy the contents of integrations/mcp-server/index.ts into supabase/functions/cerebro-mcp/index.ts.

Deploy:

supabase functions deploy cerebro-mcp --no-verify-jwt

Your MCP server is now live at:

https://YOUR_PROJECT_REF.supabase.co/functions/v1/cerebro-mcp

Build your MCP Connection URL by setting the x-brain-key header to your access key:

URL:  https://YOUR_PROJECT_REF.supabase.co/functions/v1/cerebro-mcp
Header:  x-brain-key: your-access-key

Paste both into your credential tracker.

Step 7: Connect to Your AI

Grab your MCP Connection URL from the credential tracker.

Claude Desktop

Open Claude Desktop → Settings → Connectors
Click Add custom connector
Name: Cerebro
Remote MCP server URL: paste your MCP Connection URL
Click Add

Start a new conversation and Cerebro's tools will be available.

ChatGPT

Requires a paid ChatGPT plan. Works on web at chatgpt.com.

Go to Settings → Apps & Connectors → Advanced settings → toggle Developer mode ON
Back in Apps & Connectors, click Create
Name: Cerebro
MCP endpoint URL: paste your MCP Connection URL
Authentication: No Authentication (the key is embedded in the URL)
Click Create

If ChatGPT doesn't pick up the tools on its own, tell it explicitly: "Use the search_thoughts tool to find my notes about X."

Claude Code

claude mcp add --transport http cerebro \
  https://YOUR_PROJECT_REF.supabase.co/functions/v1/cerebro-mcp \
  --header "x-brain-key: your-access-key"

Other Clients (Cursor, VS Code Copilot, Windsurf)

Option A: Remote URL. If your client supports remote MCP servers, paste the full MCP Connection URL directly.

Option B: mcp-remote bridge. For clients that only support local stdio servers, use the mcp-remote npm package to bridge the connection:

{
  "mcpServers": {
    "cerebro": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://YOUR_PROJECT_REF.supabase.co/functions/v1/cerebro-mcp",
        "--header",
        "x-brain-key:${BRAIN_KEY}"
      ],
      "env": {
        "BRAIN_KEY": "your-access-key"
      }
    }
  }
}

Step 8: Try It Out

Your AI client now has four Cerebro tools available. Here are some things to try:

What to Say	What Happens
"Save this: decided to move the launch to March 15"	Captures the thought, extracts metadata automatically
"Remember that Marcus wants to move to the platform team"	Captures with people + topics detected
"What did I capture about career changes?"	Semantic search across all thoughts
"Show me what I captured this week"	Lists recent thoughts with date filters
"How many thoughts do I have?"	Returns totals, type breakdown, top topics

Quick test — capture:

Remember this: Sarah mentioned she's thinking about leaving her job to start a consulting business

You should see a confirmation with the auto-extracted type, topics, and people.

Quick test — search:

What did I capture about Sarah?

This should return the thought above, even though you searched for "Sarah" and the thought contains "leaving her job." That's the vector similarity search at work.

Troubleshooting

Tools don't appear in Claude Desktop Check that the connector is added under Settings → Connectors and that it's toggled on for the current conversation.

401 errors on every request The access key doesn't match what's stored in Supabase Secrets. Make sure your client sends the x-brain-key header with the exact value of your MCP_ACCESS_KEY.

Search comes back empty Make sure you've captured at least one thought first. You can also try lowering the threshold: "search with threshold 0.3" casts a wider net.

Responses take several seconds The first request after a period of inactivity wakes up the Edge Function (cold start). Follow-up calls in the same session are faster. If it stays slow, check that your Supabase project region is near you.

Under the Hood

Capture flow: your text → capture_thought tool → two parallel API calls (embedding generation + LLM metadata extraction) → single row inserted into Supabase → confirmation returned.

Search flow: your query → search_thoughts tool → query gets embedded → pgvector cosine-similarity scan across all rows → results ranked by semantic closeness.

This is why "career changes" matches a note about "Sarah thinking about leaving" — the vectors encode meaning, not keywords.

Changing AI Models

Since all model calls go through OpenRouter, you can swap to a different model by editing the model identifiers in index.ts and redeploying. Browse options at openrouter.ai/models. Just keep embedding output at 1536 dimensions to stay compatible with existing data.

✅ Verification Checklist

Before moving on, confirm all of these pass:

Supabase Dashboard → Table Editor shows thoughts table with 6 columns (id, content, embedding, metadata, created_at, updated_at)
Supabase Dashboard → Database → Functions shows match_thoughts
Edge Function responds — visiting https://YOUR_PROJECT_REF.supabase.co/functions/v1/cerebro-mcp in a browser returns a page (not a 404 or 500)
Capture works — in your AI client, say "Remember this: Sarah mentioned she's thinking about leaving her job to start a consulting business" → you get a confirmation with metadata (topics, people, type)
Search works — ask "What did I capture about Sarah?" → returns the thought you just captured with a similarity score
Supabase data — Table Editor → thoughts shows at least one row with metadata.source = "mcp" and a non-null embedding

If any check fails, see the Troubleshooting section above.

Next Steps

Once your MCP server is working, you can add more ways to capture thoughts. See the Cerebro Setup Guide for the recommended order, or jump directly:

Integration	Guide	What It Does
Microsoft Teams	Teams Setup	DM or @mention a bot in Teams
Discord	Discord Setup	`/capture` and `/search` slash commands
Alexa	Alexa Setup	"Alexa, tell cerebro …" voice commands
Calendar Reminders	Reminders Setup	Auto-create O365/Google calendar events from captured dates
Daily Digest	Digest Setup	Automated daily/weekly summaries to Teams + Discord
Microsoft Graph daily ingest	Graph Ingest Setup	Once configured, your M365 mail / calendar / OneNote / OneDrive content auto-flows into Cerebro every morning

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Build Your Cerebro

What You're Building

What You Need

Services (All Free Tier)

Cost

Credential Tracker

Step 1: Create Your Supabase Project

Step 2: Set Up the Database

Enable the Vector Extension

Create the Thoughts Table

Create the Search Function

Lock Down Security

Verify

Step 3: Save Your Connection Details

Step 4: Get an OpenRouter API Key

Step 5: Create an Access Key

Step 6: Deploy the MCP Server

Install the Supabase CLI

Log In and Link

Set Your Secrets

Create and Deploy the Function

Step 7: Connect to Your AI

Claude Desktop

ChatGPT

Claude Code

Other Clients (Cursor, VS Code Copilot, Windsurf)

Step 8: Try It Out

Troubleshooting

Under the Hood

Changing AI Models

✅ Verification Checklist

Next Steps

FilesExpand file tree

01-getting-started.md

Latest commit

History

01-getting-started.md

File metadata and controls

Build Your Cerebro

What You're Building

What You Need

Services (All Free Tier)

Cost

Credential Tracker

Step 1: Create Your Supabase Project

Step 2: Set Up the Database

Enable the Vector Extension

Create the Thoughts Table

Create the Search Function

Lock Down Security

Verify

Step 3: Save Your Connection Details

Step 4: Get an OpenRouter API Key

Step 5: Create an Access Key

Step 6: Deploy the MCP Server

Install the Supabase CLI

Log In and Link

Set Your Secrets

Create and Deploy the Function

Step 7: Connect to Your AI

Claude Desktop

ChatGPT

Claude Code

Other Clients (Cursor, VS Code Copilot, Windsurf)

Step 8: Try It Out

Troubleshooting

Under the Hood

Changing AI Models

✅ Verification Checklist

Next Steps