The Comsky Orbit connector for Claude

What this connector is

An MCP server we run at mcp.orbit.comsky.ai. It sits between Claude and the Comsky Orbit API that the app itself uses, and it holds no data of its own: no database, no cache, no copy of a channel or a task. Every request it makes upstream carries your credential, so it reaches exactly what you reach and nothing more.

There is no service account behind it and no key that reads every workspace. A total compromise of the process is worth the traffic passing through it at that moment.

Connecting it

In Claude, add a custom connector pointing at https://mcp.orbit.comsky.ai/mcp. Authentication is a bearer token: either an Orbit bot token, or a sign-in through account.comsky.ai that asks you to approve what the connector may do before it is given anything.

A sign-in token is usable only while your account belongs to exactly one active workspace. Somebody in two is refused rather than guessed at — the wrong guess would put one organisation's work in front of a model answering about another, and nothing in the request could correct it.

What it can read

• Tasks — yours, the ones you delegated, and any you may see; one task in full.
• Projects — the list, one project, and its tasks.
• Channels — the list, one channel, and its messages.
• People — the directory and one member's profile.
• Search — across messages, tasks and people.
• Your day — the agenda for a given date.

What it can change

Two tools: create a task, and update one. That is the whole of it, and it is the whole of what the sign-in screen offers to approve.

Deliberately narrower than it looks. Attachments, checklist items and the task action routes all live under the same part of the API and none of them is "create or change a task", so the connector is allowed those two addresses exactly rather than that area of the product.

What it cannot do

• Pay, compensation, appraisals and performance records are out of reach. Not by the absence of a tool — by a second fence that refuses those addresses whatever the person's seniority. A manager's token cannot follow their authority into their reports' salaries.
• The audit log, administration and settings are out of reach for the same reason.
• Nothing is ever deleted. There is no delete tool and the outbound code cannot issue one; a build check fails the release if one appears.
• A read-scoped token changes nothing at all — it is held to safe HTTP methods, so a route added later is covered without anyone having to remember.

It acts as you, never above you

Orbit's own permission matrix still runs in full. The connector can never do more than you could in the app — only less, because the two fences above sit in front of it. If a task is invisible to you in Orbit, asking Claude about it returns nothing.

What happens to your data

What you ask Claude, and what the connector returns, go to Anthropic and are handled under their terms. On our side the connector stores nothing: no transcript, no copy of a channel, no record of what was asked. Its logs name the tool that ran and the arguments it took by name, never by value.

A write is recorded in the workspace's own audit trail, exactly as if you had done it in the app.

How Comhard Technologies handles personal information generally is in the Privacy Policy.

Rate limits

Per credential: sixty requests a minute, four at a time, and a budget on how many upstream calls one tool may make. Writes are capped separately at ten a minute. The workspace has its own shared ceiling on top.

Support

Write to [email protected], or [email protected] for anything that looks like a security problem.