What is MCP?
Model Context Protocol is an open standard that lets an AI assistant use the tools of another system. With it, your assistant reads and acts inside Unbound OS.
In practice you ask your assistant something like "log the call with this lead and set a follow up for Thursday", and it calls the workspace tools that do that, with your permission.
How many tools are there?
There are 438 names in the catalog; 423 work and 15 are retired names that only point to their replacement. 128 read and 310 write.
| Module | Tools |
|---|---|
| Clients and portal | 82 |
| Core | 68 |
| CRM and pipeline | 55 |
| Calendar and booking | 51 |
| Documents and proposals | 47 |
| Finance | 40 |
| Library and data rooms | 35 |
| Social | 20 |
| Forms | 17 |
| Skills | 13 |
| KOL and media | 10 |
| Campaigns | 0 |
Per module counts include the 15 retired names. What each agent sees is less: the catalog is filtered by the workspace’s active modules, the person’s permissions and the key’s scope.
Campaigns has no MCP tools yet. That is why we never say "every module".
How do I connect it?
Two ways: by OAuth from your assistant, or with an API key. Both give exactly the same permissions.
- 01OAuth. In your assistant, add an MCP connector with the server address, https://unboundoperators.app/api/mcp. The server supports OAuth 2.0 with dynamic client registration, so the assistant registers itself and asks you to sign in and authorize. Claude is the tested client; any MCP client that supports OAuth works.
- 02API key. Create a key in the workspace and use it with clients that do not do OAuth. The key inherits its owner’s permissions and can be narrowed to fewer permission groups.
- 03Service accounts. For agents that run on their own, create a service account: it cannot sign in, starts with no permissions, reaches the workspace only through keys and does not take a seat on the bill.
What can an agent do?
Whatever the person or service account that connected it can do, never more, and only inside one workspace.
- Permissions are per person and per permission group. An agent connected by someone without Finance access sees no invoices.
- A key can be narrowed below its owner’s permissions.
- Every query is scoped to the workspace that makes it, exactly as it is for a person.
Do actions wait for approval?
Your call. By default, an action the person is allowed to take runs directly. Switch off "approve my own MCP actions" and everything waits for review.
A key narrowed to fewer permissions also sends anything outside its scope to the review queue. Actions in review are approved or denied from the workspace.
Which model do I use, and what does it cost?
The one you already pay for. Unbound OS does not resell AI or mark it up: you connect your assistant and pay its provider directly.
MCP access comes with Core at no extra cost. Each module’s tools appear when that module is on for that person.
Do other tools have MCP too?
Yes. HubSpot, ClickUp, monday.com, Notion, PandaDoc, HoneyBook and Bonsai run official MCP servers. Dubsado and SuiteDash do not.
The difference is not having MCP. It is that in Unbound OS one connector reaches the whole operation, from proposal to invoice, on the same record per client, with each person’s permissions and no AI markup. The comparisons are on compare.
What does it look like in practice?
You ask for a task with several steps and the agent calls the tools in order, each step recorded in the workspace.
For example: "log the call with the new lead, draft a proposal from our template and set a follow up for Thursday". The agent calls, in order:
- 01log_relation_activity: the call lands on the relation.
- 02create_proposal: the draft proposal, from the template.
- 03create_followup: the follow up for Thursday.
Frequently asked questions
Does it work with ChatGPT?
It works with any MCP client that supports OAuth or API keys. Claude is the client we test with; there is no code specific to other assistants.
Are agent keys billed?
No. Service accounts and keys do not take a seat on the bill.
Can an agent see another company’s data?
No. Every query is scoped to the workspace that makes it, and the database has row level security on every table.
Can I see what an agent did?
Yes. What an agent does lands on the same records as what a person does, and what goes to review waits for your approval.