infera-mcp-server

Lets an AI assistant like Claude or ChatGPT access your company's internal database and give you insights — with controlled access, only to what it needs.

Architecture

Related MCP server: Alma Atlas

Company records it can pull up

• Customers — who they are, when they joined, and whether they're active, churned, or a prospect.

• Subscriptions & plans — what each customer is subscribed to, and every upgrade, downgrade, or cancellation along the way.

• Sales deals — every closed deal, won or lost.

• Sales pipeline — every open opportunity still being worked.

Available MCP tools

Sample analytical questions

• What's our monthly recurring revenue right now, and is our cash flow trending up or down?

• Who are our top customers, and which ones are we at risk of losing?

• How much revenue is sitting in open deals, and how likely are they to close?

• Are we losing customers faster than we're signing new ones this month?

• What's our sales win rate, and how big are deals closing on average?

Potential integrations

Not yet built — natural next steps beyond read-only reporting:

• Calendar — read schedules, create meeting events (book a call with an at-risk customer, schedule the board update).

• Email — read inbox, send new emails, reply to threads (follow up a stalled deal, send the monthly digest).

• Slack — same as email: read channels/DMs, send messages, reply in-thread.

• CRM — another database, but for customer/deal records (HubSpot/Salesforce) — read and write, alongside the analytical warehouse.

Once wired in, this unlocks questions like: "How many leads do we have, how many have we contacted, how many replied, and what platform is each one in (email vs. Slack vs. CRM)?"

Setup (Docker)

Run MCP Server and Postgres Database Engine locally using Docker

cp .env.example .env
docker compose up -d --build
docker compose run --rm mcp python -m warehouse.seed.load

Once the container's running, connect it to Claude Code by typing this on terminal:

claude mcp add infera --transport http http://localhost:8000 --header "Authorization: Bearer <MCP_API_KEY>"

Use the same MCP_API_KEY value set in .env.