Connect Claude, Codex, GitHub Copilot or Microsoft 365 Copilot to Skills Workflow
Add Skills Workflow to the AI tool you work in. Once it is connected, the tool can create status pages, reports and screens for your agency and publish them to your tenant. The connection is the same in every tool: one address and three headers.
Prerequisites
Ask the Skills Workflow team for the connection details of your tenant. You receive four values:
| Value | Where it goes |
|---|---|
| Your tenant's API address | The connector's address |
| App ID | The x-app-id header |
| App key | The x-app-key header |
| Tenant ID | The x-tenant-id header |
The connector's address is your tenant's API address followed by /mcp/client-portal. For a tenant whose API address is https://apiv2-demo-test-we.skillsworkflow.com, it is:
https://apiv2-demo-test-we.skillsworkflow.com/mcp/client-portal
With the details come two skills, client-portal and app-page. They give the AI tool the rules of what it publishes: how a page reads your data, what it may load and how it is published. Add them to your tool before you ask for anything; in Claude, under Settings > Capabilities > Skills.
Add the connector
Claude
Claude Code
ChatGPT / Codex
GitHub Copilot
Microsoft 365 Copilot
Claude on the web and Claude Desktop share connectors. On Team and Enterprise plans, an owner of the Claude organisation adds the connector for everyone.
-
Open Settings > Connectors and select Add custom connector.
-
In the first field, type a name for the connector, for example
Skills Workflow. It is the name shown in the connectors list. -
In the second field, type the connector's address and select Continue.
-
Claude checks the server. Connecting to the server answers
401and the sign-in checks answer404: that is expected, since the connector signs in with headers. When Server found appears, select Continue.
Claude finds the Skills Workflow server at the connector's address. -
Under Authentication, select No sign-in.
-
Under Request headers, select Add header three times and fill in
x-app-key,x-app-idandx-tenant-id, each with the value the Skills Workflow team sent. Leave Required selected on each one. -
Select Add.
No sign-in, with the three request headers of the Skills Workflow connector.
Claude stores the header values and never shows them again. To change one, remove the connector and add it again.
Add the connector from a terminal:
claude mcp add --transport http skills-workflow https://apiv2-demo-test-we.skillsworkflow.com/mcp/client-portal \
--header "x-app-id: <App ID>" \
--header "x-app-key: <App key>" \
--header "x-tenant-id: <Tenant ID>"
Replace the address with your tenant's and each <...> with the value the Skills Workflow team sent.
In the ChatGPT app:
-
Open Plugins and select the MCPs tab.
-
Select Add and add an MCP server.
-
In URL, type the connector's address.
-
Leave Bearer token env var empty.
-
Under Headers, select Add header for each credential and fill in
x-app-id,x-tenant-idandx-app-key, each with the value the Skills Workflow team sent. -
Select Save.
The Skills Workflow MCP server in the ChatGPT app, with the connector's address and its three headers.
The Codex CLI and the Codex IDE extension read the same servers from ~/.codex/config.toml. To add the connector there instead:
[mcp_servers.skills-workflow]
url = "https://apiv2-demo-test-we.skillsworkflow.com/mcp/client-portal"
http_headers = { "x-app-id" = "<App ID>", "x-tenant-id" = "<Tenant ID>" }
env_http_headers = { "x-app-key" = "SW_APP_KEY" }
Replace the address with your tenant's and each <...> with the value the Skills Workflow team sent. The app key is read from the SW_APP_KEY environment variable; set it to the app key before starting Codex.
In Visual Studio Code, add the connector to mcp.json, either in your user profile or in .vscode/mcp.json of a project. Run MCP: Open User Configuration from the Command Palette for the user file.
{
"inputs": [
{ "type": "promptString", "id": "sw-app-id", "description": "Skills Workflow App ID" },
{ "type": "promptString", "id": "sw-app-key", "description": "Skills Workflow App key", "password": true },
{ "type": "promptString", "id": "sw-tenant-id", "description": "Skills Workflow Tenant ID" }
],
"servers": {
"skills-workflow": {
"type": "http",
"url": "https://apiv2-demo-test-we.skillsworkflow.com/mcp/client-portal",
"headers": {
"x-app-id": "${input:sw-app-id}",
"x-app-key": "${input:sw-app-key}",
"x-tenant-id": "${input:sw-tenant-id}"
}
}
}
}
Replace the address with your tenant's. The first time the server starts, Visual Studio Code asks for the three values and stores them. Do not write the values into the file of a project you commit.
Create a Skills Workflow agent in Copilot Studio for your agency's Microsoft 365 Copilot or Teams users. The agent connects to your tenant through the Skills Workflow MCP server.
Prerequisites
Ask Skills Workflow for your tenant's API address, App ID, App key, Tenant ID, and the client-portal and app-page skills. You need access to Copilot Studio and permission to create custom connectors in the same Power Platform environment.
Use a Power Platform environment with Dataverse. Keep the App key with authorised publishers: it permits publishing and replacing pages across the tenant as its Skills Workflow administrator.
Create the custom connector
- Open Power Apps, select the environment, and open Custom connectors → New custom connector → Create from blank. Name it
Skills Workflow MCP. - Enable Swagger editor. Download the OpenAPI template, replace
YOUR-TENANT-API-HOSTwith your tenant's API hostname (withouthttps://or a path), and paste the definition into the editor. The MCP path is/mcp/client-portal. - Disable Swagger editor and check Security: authentication API Key, parameter name
x-app-key, location Header. Select Create connector.
- Open Test → New connection, enter the App key in the protected
api_keyfield, and select Create. This creates the connection; a Connected status alone does not validate the MCP tools.
Create a standard agent
- In Copilot Studio, select the same environment. Open Home → Other ways to build → Build with standard orchestration → Agent. Enter a name and select Create.
- Open Add tool, search for
Skills Workflow, and select Skills Workflow MCP under Model Context Protocol. Select the connection you created and choose Add. - Open the added tool and fill in Inputs:
x-app-idwith App ID andx-tenant-idwith Tenant ID. Select Save. Both are headers sent with the App key from the connection. Fill them for each agent; they are not copied automatically from another agent.
Test the MCP connection
Start a new test session and ask which Skills Workflow tools are available. Compare the list with the eight tools in the Check the connection section below.
Then ask the agent to call get_portal_url for the hypothetical client mcp-connection-check and portal connection-check, without creating or modifying anything. Inspect the activity trace: the tool must finish with Completed and return your tenant's URL. This tool returns an address even when no page exists there; it does not publish or verify the page.
A textual list of tools is not enough to validate the connection. If the editor reports Connector request failed, check the three header values and saved inputs. Use the actual tool result in the test trace to confirm runtime behaviour.
Prepare access for key users
Before asking the agent to publish content, obtain and configure the official client-portal and app-page rules with Skills Workflow. The connection test above does not publish the agent or verify its skills.
For organisational rollout, follow Microsoft's Teams and Microsoft Copilot publishing instructions. Have IT review licensing, connection identity and the intended group of authorised publishers before making the agent available. People opening published pages sign in as themselves.
Trial and credits
A Copilot Studio trial can allow testing a standard agent. Other agent types can require Copilot Credits for building and testing. If you see EnforcementUsageCredits, ask your administrator to check Licensing → Copilot Studio → Manage Copilot Credits. See Microsoft's capacity guidance.
Screens inside Skills Workflow are built as a small code project. Build them in a tool that works on code: Claude Code, Codex or GitHub Copilot.
Check the connection
Start a new conversation and ask:
Which Skills Workflow tools do you have?
The answer lists the connector's eight tools:
| Tool | What it does |
|---|---|
get_client_portal | Reads a published client page |
create_client_portal | Publishes a new client page |
update_client_portal | Replaces a published client page |
get_portal_url | Gives the address of a client page |
get_app_page | Reads a screen inside Skills Workflow and its versions |
publish_app_page_version | Publishes a new version of a screen |
set_app_page_live_version | Chooses the version of a screen users get |
get_app_page_url | Gives the address of a screen |
Nothing is read from or written to your tenant by this question. If the tool says it has no Skills Workflow tools, or reports an error, one of the values is wrong; compare the address and each header with what the Skills Workflow team sent.
Rules and behaviour
- The credentials work only on your own tenant's API address. Any other address refuses them.
- The connector publishes as your tenant's Skills Workflow administrator. Anyone with the app key can create and replace pages for every client of your tenant, so keep it with the few people who publish. If the key reaches someone it should not, tell the Skills Workflow team.
- The connector reaches your tenant's file system only. It publishes, reads back and gives the address of what it publishes; it does not change jobs, projects or any other record.
- People who open what was published sign in as themselves. The administrator identity is used to publish, never to show data.