Skip to main content

Vault

Every artifact a run produces — extracted datasets, CSVs, generated files — lands in the Vault. Artifacts stay linked to the run that made them, so you can always trace a number back to the job that produced it. Download them from the dashboard, or over the API: a run’s detail response includes each artifact with a signed download_url valid for one hour.

Connections

Beta.
Connect your own database and the agent can query it as part of a run — join your CRM data against freshly scraped leads, check which prospects are already customers, or analyse a table in plain language. Supported: PostgreSQL, MySQL, BigQuery.
Security specifics worth knowing:
  • The credential goes straight into an encrypted vault. It is never returned by the API — listing connections returns configuration and status only.
  • Queries run from a worker with a static egress IP, so you can allowlist it on your side.
  • ManyPI records whether the connection has been verified read-only. Give it a read-only user; there is no reason for it to have anything more.
Test a connection any time with POST /api/connections/{id}/test.

MCP connectors

Connect third-party MCP servers to your workspace and their tools become available to the ManyPI agent inside a run. Credentials are encrypted on the way in and never returned. Manage them in Settings → Integrations, or at /api/mcp-connectors. ManyPI ships a catalogue of presets for common servers, and OAuth-based servers are authorised through a browser handshake.
This is the inbound direction: other people’s tools inside ManyPI. For the outbound direction — ManyPI’s tools inside Claude, Cursor or ChatGPT — see the MCP server tab.

Studio

Pro and Business plans.
The visualization Studio turns a dataset — a scraper run, an artifact, a database query result — into charts you can save, share by link, and embed. Charts also render inline in chat when the agent decides a picture is clearer than a table.