The BrainFlow MCP server connects Claude Desktop, Cursor, and GitHub Copilot to your team's shared email memory. Ask questions in plain English. Get sourced answers in seconds. No SQL required.
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external data sources. Instead of copying and pasting context into a chat window, the AI queries your data directly — in real time, with your permission, under your control.
BrainFlow implements MCP over SSE (Server-Sent Events). Your AI assistant sends a natural language question, BrainFlow translates it to SQL, runs it against your team's email database, and returns a sourced answer with links to the original threads.
Natural language → SQL
Ask “What did Acme Corp say about pricing?” The AI translates your question to a query, BrainFlow executes it, and returns a sourced answer.
Read-only by design
The server only accepts SELECT queries. It cannot write, delete, or modify anything. This is enforced at the code level — not by policy.
Company-scoped isolation
Every query is filtered to your company. Even if two teams use BrainFlow, their data never mixes. API keys are hashed with SHA-256 and scoped per company.
“What's the average reply time for client Acme Corp?”
Returns avg reply time, fastest/slowest replies, and who replies fastest on the account.
“How many offers did we send last week?”
Lists every offer with value, date, client, and current status.
“Which deals are stuck longer than 14 days?”
Surfaces threads with no reply in the last 14 days, sorted by deal value.
“Who sent the most emails last month?”
Ranked list with counts, peak days, and thread breakdown.
“Show me unresolved high-priority tasks.”
Extracts commitments with deadlines and flags overdue items.
“What did we commit to Beta Inc in March?”
Temporal search across months with direct thread links.
“Which emails mention the Alpha contract?”
Semantic search across thread content with excerpt highlights.
“Find the latest pricing document.”
Searches attachments and filenames, returns sender, date, and thread link.
“How many emails did we receive this week?”
Volume breakdown by sender type, busiest day, and most active thread.
“Which clients emailed us about GDPR?”
Topic-based semantic search with client names and thread counts.
Sign in to your BrainFlow dashboard, go to API Keys, and clickGenerate key. Name it (e.g., “Claude Desktop”) and copy the key — you will only see it once.
In Claude Desktop, go to Settings → MCP Servers → Add Custom Server. Set the name to BrainFlow and the URL to your endpoint with the API key as a query parameter:
Name: BrainFlow
URL: https://brain-flow.ai/mcp/sse?api_key=bf_mcp_YOUR_KEY_HEREReplace YOUR_KEY_HERE with the key you generated in Step 1.
Restart Claude Desktop. You should see the BrainFlow tools available in the chat interface. Try your first query:
“What did we discuss with Acme Corp last month?”
Claude will call the BrainFlow MCP server, run the query, and return a sourced answer with direct links to the original email threads.
GitHub Copilot in VS Code also supports MCP. This means you can ask questions about client deals, project commitments, and team decisions without leaving your editor — perfect for context-switching between coding and client calls.
{
"mcp": {
"inputs": [],
"servers": {
"brainflow": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://brain-flow.ai/mcp/sse",
"--header",
"x-api-key:bf_mcp_YOUR_KEY_HERE"
]
}
}
}
}Tip: If your client does not support headers, use the query param URL: https://brain-flow.ai/mcp/sse?api_key=bf_mcp_YOUR_KEY_HERE
Cursor supports MCP natively. Add BrainFlow to your Cursor composer and ask questions about requirements, deadlines, and client feedback while you code.
Open Cursor Settings
Cmd/Ctrl + , → MCP
Add MCP Server
Paste the BrainFlow config
Query in Composer
Ask about deals while coding
Permanently read-only
Only SELECT, EXPLAIN, SHOW, DESCRIBE, and WITH queries are allowed. No UPDATE, DELETE, INSERT, DROP, or ALTER — ever.
API key authentication
Every request requires a valid API key. Keys are SHA-256 hashed in the database. Plaintext keys are never stored.
Company isolation
Every query is scoped to your company. The WHERE clause is injected automatically. No cross-tenant data access is possible.
Audit trail
Every key usage is tracked with last_used_at timestamps. Revoke keys instantly from the dashboard if a device is lost.
It is a Model Context Protocol implementation that lets Claude, Cursor, GitHub Copilot, and other MCP-compatible assistants query your team's shared email memory via natural language. The assistant sends a question, BrainFlow translates it to SQL, and returns a sourced answer.
Yes. The query validator rejects anything that is not a SELECT, EXPLAIN, SHOW, DESCRIBE, or WITH statement. Write keywords (INSERT, UPDATE, DELETE, DROP, ALTER, CREATE, TRUNCATE, GRANT) are blocked with a hardcoded allow-list. There is no environment variable or setting to disable this.
Any assistant that implements MCP: Claude Desktop (Anthropic), Cursor, GitHub Copilot (VS Code), and any other tool that supports the Model Context Protocol over SSE.
No. The MCP server requires an active BrainFlow subscription because it queries your team's shared memory database. There must be email data in BrainFlow for the server to return meaningful answers.
Go to your BrainFlow dashboard → API Keys. Click the revoke button next to any key. The key is deactivated immediately and will be rejected on the next request.
Revoke the lost key from your dashboard and generate a new one. Because keys are SHA-256 hashed, we cannot recover the plaintext. You must create a new key.
Yes. Each API key is subject to fair-use rate limits based on your plan. Startup plans include higher throughput than Founder plans. Contact us if you need enterprise-level quotas.
Self-hosting is available on Corporate plans. Contact us for deployment documentation and infrastructure requirements. All self-hosted instances still require a valid BrainFlow API key for authentication.
Generate your API key in the BrainFlow dashboard and connect your AI assistant in under 5 minutes.