Work Your Data From Claude, ChatGPT, or Your Terminal
Connect Claude or ChatGPT to your workspaces with a hosted MCP server, or install the InfoLobby Agent Skill into Claude Code, Cursor, or any local agent.
Give an AI assistant real access to your records, scoped to the workspaces you pick and revocable in one click.
Why this matters
AI assistants are useful right up to the moment they need your actual data. Copying records into a chat window is slow, goes stale immediately, and leaks more than you meant to share. Handing over a raw API key is worse: it has no scope, no expiry, and nothing to revoke short of deleting it.
InfoLobby runs a remote MCP server at infolobby.com/mcp. Add it as a connector in Claude or ChatGPT, sign in, choose which workspaces to share, and the assistant can find records, answer questions across tables, create and update rows, read and post comments, and handle attachments. For local agents like Claude Code, Cursor, Gemini CLI, or Codex, install the InfoLobby Agent Skill instead: same capabilities, a fraction of the context cost.
Two paths, because the sandbox decides which one works.
A hosted assistant (Claude on the web, ChatGPT) cannot reliably make outbound calls from its sandbox, so it needs the MCP connector. A local agent already has network access, and an MCP server's tool definitions sit in the model's context on every single turn whether they are used or not. A skill costs about a hundred tokens until the moment it triggers. So: connector for hosted assistants, skill for anything running on your machine.
- Claude on the web or ChatGPT: add the connector at infolobby.com/mcp
- Claude Code, Cursor, Gemini CLI, Codex: install the Agent Skill
- Both surfaces run on the same public API, permissions, and rate limits
What changes for the team
- Assistants read and write live records instead of a pasted snapshot
- No API key to paste for the connector. You sign in and approve
- Pick exactly which workspaces an assistant can see
- The connection acts as you, so your roles and field permissions still apply
- Revoke any connected app instantly from Account, Connected Apps
Feature details
- Remote MCP server at infolobby.com/mcp for Claude and ChatGPT connectors
- Agent Skill install for Claude Code, Cursor, Gemini CLI, Codex, Goose, and other Agent Skills clients
- Discovery tools: list workspaces and tables, read a table's field schema, list saved views
- Record tools: query with filters and sorting, read, create, update, delete
- Comments and attachments: read a thread, post a comment, upload or remove files
- Flow authoring for account owners, with new flows created switched off until you enable them
How teams usually start
- In Claude or ChatGPT, open Settings, Connectors, add a custom connector, and paste https://infolobby.com/mcp.
- Sign in to InfoLobby, tick the workspaces you want to share, and approve. Review what is listed on the consent screen before you do.
- For a local agent, run the one-line installer on the Agent Skill page instead, and give it an account or personal API key the first time it runs.
FAQ
How do I connect Claude or ChatGPT to InfoLobby?
Open Settings, Connectors in Claude or ChatGPT, add a custom connector, and paste https://infolobby.com/mcp. You sign in to InfoLobby, choose which workspaces to share, and approve. There is no API key to paste.
What can the assistant actually do with my data?
It can list workspaces and tables, read a table's schema and saved views, query and read records, create, update and delete them, read and post comments, and handle file attachments. Account owners can also have it author flows, which start switched off. It cannot create or restructure workspaces and tables, manage members, or touch billing.
Can it see data I am not allowed to see?
No. The connection acts as you. Your workspace roles and field permissions apply unchanged, and it is limited further to the workspaces you ticked when you authorized it.
How do I revoke access?
Go to Account, Connected Apps. Every authorization is listed with its access level, the workspaces it can see, and when it was last used. Revoke takes effect immediately.
Should I use the MCP connector or the Agent Skill?
Use the connector for Claude on the web and ChatGPT, whose sandboxes cannot make the outbound calls a skill needs. Use the Agent Skill for local agents like Claude Code, Cursor, Gemini CLI, or Codex: an MCP server's tool definitions occupy context on every turn, while a skill loads only when it is needed.
Does it work with agents other than Claude?
Yes. ChatGPT connects through the same MCP server, and the skill uses the open Agent Skills format, so Cursor, Gemini CLI, Codex, Goose, OpenCode and other compatible clients can use it. Scripted clients can also send an existing InfoLobby API key as a bearer token against the MCP endpoint.
Want this feature in a system your team can actually run?
Start with the workflow this feature supports, then add the records, views, tasks, comments, and automations around it.