For teams that already have an agent

Keep your agent. Connect it to trusted context.

Connect Claude Code, Codex or another compatible agent to your business and application model over MCP. It can read who you serve, what they expect and how the system is designed, while you keep working in your existing tools.

Remote MCP serverhttps://agents.cisely.ai/mcp

What the agent gains

A shared source of truth for people and agents.

A prompt can describe the immediate task, but it rarely contains the business context and system design behind it. Connected to Cisely, your agent can read the same structured model as your team.

Who you serve, and what they expect
Personas, operating contexts, stakeholders, goals and expectations tell the agent who a change is for and what outcome it should support.
Why the company exists, and what it believes
Mission, beliefs, strategy, initiatives and metrics give each task the wider business rationale a short prompt cannot carry.
How the system is designed
Entities, commands, queries, policies and routes expose the intended design without forcing the agent to infer it from code.
Product documentation
The agent can search Cisely documentation directly when it needs help with the platform.
A model that improves with use
With your permissions, the agent can record decisions, add expectations and propose corrections as the work reveals new information.

Setup

Connect with one URL and a browser sign-in.

The Cisely MCP server is hosted, so there is nothing to install or run locally. Add the address to your agent, sign in through the browser and the Cisely tools become available without a long-lived API token in a config file.

Claude Code
claude mcp add --transport http cisely https://agents.cisely.ai/mcp — then /mcp inside Claude Code to finish the browser authorization.
Claude, desktop and web
Settings → Connectors → Add custom connector, paste the URL, and sign in.
Codex
A remote MCP server in ~/.codex/config.toml, or scoped to one project — then codex mcp list to confirm it loaded.
Anything else that speaks MCP
The pattern is the same everywhere: register a remote HTTP MCP server at the URL above and complete the sign-in.

What holds, whoever is asking

The agent works within your existing permissions.

Every call runs as the signed-in user through the same permission-checked operations used by the web application. Connecting an agent does not create a separate path around access control.

The same permissions
Tenancy, authorization and entitlements apply to agent calls just as they do to your own actions.
No standing credential
Authorization is a browser sign-in rather than a long-lived token living in a config file on a laptop.
Inside your workspace
The agent can reach only the model and resources available to your account in your workspace.
Included with your account
MCP access comes with Cisely rather than as a separate product. Your subscription tier sets the rate limits.

Against the built-in agent

Choose the level of guidance you need.

Both options use the same Cisely model and permissions. The MCP connection gives your agent access to platform operations. Paroksh also brings Cisely’s draft-first workflow, decision gates and built-in checks for missing rationale.

AspectParokshYour own agent
Where it runsInside CiselyWherever your agent runs
Sees the screen you are onYesNo
The draft-and-activate loopBuilt inYour agent’s own behaviour
Blocking questions, capped ask formsBuilt inNot provided
Session handover and rechargeBuilt inYour agent’s own context handling
Model usage bills toCisely, or your providerYour provider

What is enforced for both: every call runs as the signed-in person, so permissions, tenancy and entitlements apply identically. And running both is fine — they read and write the same model, which is the point of it being a model.

Two different MCP servers, worth separating

One lets an agent read your model. The other exposes what you built.

This page is about the first: https://agents.cisely.ai/mcp, which gives your agent the context. Applications built on Cisely also publish an MCP endpoint of their own, so agents can call the system you shipped under the same rules as everything else. They are different servers answering different questions, and it is worth not confusing them.