# Claude Cowork integrations, and the open source layer that governs them

> Claude Cowork connects to your apps through MCP connectors, and Kortix is open source: the AI Operating System where a team sets each tool to Allow, Ask or Block. This page covers what connects, how a connector is authorized, and the governance gap a team hits.

URL: https://opensourceclaudecowork.com/claude-cowork-integrations

_Compare_

## Connector governance, Kortix and Claude Cowork side by side

Kortix leads the table; the Claude Cowork column states only what Anthropic's own help pages document.

| | Kortix (recommended) | Claude Cowork |
| --- | --- | --- |
| What connects | 3,000+ apps in a click, plus MCP, OpenAPI, GraphQL and raw HTTP. | The Connectors Directory and custom MCP servers. |
| Who can use it | Per-resource permissions for people and agents. Roles, groups, and an audit trail. | An Owner enables it for the org; each person authenticates separately. |
| The rule on a call | Set each tool to Allow, Ask or Block. | Owners set Always allow, Needs approval, or Blocked. |
| Approvals | Approval gates you set. Off until you set them. | Deletion needs explicit permission; computer use runs without permission checks. |
| Credentials and access | Connector credentials are brokered server-side and never enter the machine. | Claude inherits each person's permissions from the connected service. |

_Quickstart_

## Connectors are files a team can review

`kortix.yaml`

```yaml
# kortix.yaml
agents:
  support-triage:
    file: agents/support-triage.md   # what it does
    connectors: all                  # what it may reach
```

Source: https://kortix.com/docs

The connectors an agent may reach live in kortix.yaml, so a team reads and diffs them like any other code.

## What Claude Cowork connects to

Claude Cowork is Anthropic's agent for knowledge work. You describe an outcome, and it runs multi-step tasks on your behalf inside Claude's app on desktop, web and mobile. It reaches your tools through connectors, which Anthropic documents as the way Claude accesses your apps and services, retrieves your data, and takes actions inside them ([Anthropic's connector guide](https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities)).

The Connectors Directory holds the prebuilt connectors, and Anthropic's own examples include Google Drive to search your files, GitHub to work with repositories, Linear to create issues, and Slack to send messages. Beyond the directory, Claude supports custom connectors over a remote MCP server on free, Pro, Max, Team and Enterprise plans, with free accounts limited to one custom connector. Cowork also produces Excel spreadsheets and PowerPoint decks that you can keep editing in Claude for Excel and PowerPoint ([Get started with Claude Cowork](https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork)).

## How a connector is authorized and scoped

Authorization is per person and per connector. Anthropic's guide says Claude inherits each person's permissions from the connected service: if someone cannot open a file in the source app, the connector cannot reach it either. On Team and Enterprise plans, an Owner or Primary Owner enables a connector for the organization first, but each member still authenticates on their own before they can use it. Enterprise-managed auth, in beta, authorizes a connector once for the whole organization.

Owners can also restrict what a connector may do org-wide, allowing reads while blocking writes, and each person still needs the matching permission in the source system. That set of controls is documented and real. What it does not answer is the question a team with several agents has to answer.

## The governance gap: which agent may touch which tool

The controls Anthropic documents are set at the connector and the organization. A team that runs agents across shared systems needs a finer unit: which agent may touch which tool, with which arguments, and where the credential that reaches the tool actually sits. Cowork's model covers the plan you approve and the deletion step, and Anthropic's own safe-use guidance notes that computer use runs without the permission checks that gate other Cowork tools ([Use Claude Cowork safely](https://support.claude.com/en/articles/13364135-use-claude-cowork-safely)). The per-agent question stays open. [Claude Cowork for teams](/claude-cowork-for-teams) covers the admin controls Cowork does provide.

## How Kortix governs connector access

Kortix is the open-source AI Operating System, and its connector layer is a file in a git repo you own. Connectors reach "3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP." Connector credentials are brokered server-side and never enter the machine. The rule on a call is the team's to set: set each tool to Allow, Ask or Block, down to the arguments of each call. An Ask holds the call until a person approves it, then the agent resumes. Approval gates you set. Off until you set them.

Permissions are per resource, for people and agents, with roles, groups, and an audit trail. The connector configuration travels with the project, so a reviewer reads the same file the agent runs from. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free ([the self-hosting guide](/self-hosting)). The source is public at [Kortix on GitHub](https://github.com/kortix-ai/suna).

_FAQ_

## Claude Cowork integrations, the questions teams ask

### Which Claude plans include connectors?

Connectors are available on free, Pro, Max, Team and Enterprise plans across Claude, Cowork and Claude Desktop, and free accounts are limited to one custom connector. On Team and Enterprise plans an Owner enables a connector for the organization, and each person still authenticates.

### Can Kortix reach an app that has no ready-made connector?

Yes. Kortix reaches 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP, so a team connects a custom service through MCP or a REST, GraphQL or OpenAPI definition.

### Where do connector credentials live in Kortix?

Connector credentials are brokered server-side and never enter the machine. Secrets are encrypted at rest with a key per project.

## Set the rule on every connector call

Start with one connected app, in one repo you own.

[Try Kortix](https://kortix.com) · [Compare the tools](https://opensourceclaudecowork.com/claude-cowork-alternatives)