Start in the cloud
1
Sign in and add an LLM key
Create an org at corebasehq.com. CoreBase is bring-your-own-key — in Settings → LLM keys, paste an Anthropic, OpenAI, Gemini, Groq, Mistral, Azure, OpenRouter, or Ollama key. Nothing runs without one. See LLM keys.
2
Connect a source
In Sources, add what your agent will work on:
- Database — paste a PostgreSQL DSN (cloud-direct), or register a CoreMCP agent for SQL Server / Firebird.
- API — give a REST OpenAPI URL or a GraphQL endpoint.
- SaaS app — connect Jira, Gmail, GitHub, and 50+ more via OAuth. See Connectors.
3
Ask your data
Open Chat and ask a question to confirm the source works:
How many customers signed up last month and what’s the average order value?The assistant explores the schema, writes and runs the queries itself — read-only, scoped to the sources you picked.
4
Ship it
Two ways to put it in front of your users:
- API — create a token under API Tokens and call the OpenAI-compatible chat endpoint from your own app (or from n8n/Zapier with
stream: false). - Widget — embed the assistant in your product with a server-signed JWT; each end-user gets their own isolated, rate-limited session.
Reach a database without an API, or one behind a firewall (CoreMCP)
For SQL Server, Firebird, or any database behind a firewall, run the open-source CoreMCP agent inside your network. It holds the credentials, runs queries locally, and dials outbound only.1
Install CoreMCP
docker pull y11t0/coremcp:latest. From source? go build ./cmd/coremcp. See Install.2
Configure a source
Create See Connect a data source for
coremcp.yaml with a SELECT-only database user:mssql, rest, and graphql DSNs; full schema in Configuration.3
Connect it to CoreBase
In the panel, Sources → Add agent, register your CoreMCP agent. Its tables surface as a source your agents can use — same governance, same Unified Context.Working directly in Claude Desktop or Cursor instead? Point the client at the binary over stdio — see MCP clients. Raw rows stay on your network.
Next steps
Embeddable widget
Ship the assistant inside your own product.
Governance
Read-only defaults, write opt-in, source scoping.
Unified Context
Annotations, glossary, Query Memory.
Developer API
OpenAI-compatible
POST /api/v1/chat/completions.