Cisely has an agent built in — Paroksh — and for most people that is the whole story. This page is for the case where you already have an AI agent and a workflow you like, and would rather bring it than switch.
The MCP connector lets an outside agent read and write your Cisely model on your behalf. It is a remote server, so there is nothing to download: point your agent at one URL and sign in through the browser.
https://agents.cisely.ai/mcpAuthentication is a browser sign-in — the agent acts as you, inside your workspace, through the same permission-checked operations you would use by hand. There is no API token or secret to paste into a config file.
Claude Code
claude mcp add --transport http cisely https://agents.cisely.ai/mcpRun /mcp inside Claude Code to finish the browser authorization, and claude mcp list to confirm it is connected.
Claude (desktop and web)
Add it as a custom connector: Settings → Connectors → Add custom connector, paste https://agents.cisely.ai/mcp, and complete the sign-in.
ChatGPT Codex
Add a remote MCP server to ~/.codex/config.toml (or .codex/config.toml to scope it to one project):
[mcp_servers.cisely]
url = "https://agents.cisely.ai/mcp"Then confirm it loaded with codex mcp list.
Any other MCP client
The pattern is the same everywhere: register a remote / HTTP MCP server with the URL above and complete the browser sign-in. Your agent then gains a set of Cisely tools — reading and writing your model, and searching the documentation you are reading now.
How this differs from Paroksh
Worth understanding before you choose, because they are not the same thing wearing different clothes.
| Paroksh | Your own agent over MCP | |
|---|---|---|
| Where it runs | inside Cisely | wherever your agent runs |
| Sees the screen you are on | yes | no |
| The drafts-and-activate loop | built in | your agent's own behaviour |
| Blocking questions, capped ask forms | built in | not provided |
| Session handover and recharge | built in | your agent's own context handling |
| Bills to | Cisely, or your provider if you choose one | your provider |
The MCP surface gives an agent the operations. The guardrails described in Working with Paroksh — the draft-first loop, the blocking decision points, the refusal to proceed on an unjustified change — are Paroksh's behaviour, not properties of the connector.
What is enforced for both: every call runs as the signed-in person, so permissions, tenancy and entitlements apply identically. An outside agent is not an escape hatch around them.
Both at once is fine. They are two clients of the same platform, and the model is the shared source of truth.