MCP

The Convert Model Context Protocol (MCP) server connects compatible AI clients to your Convert account and documentation.
It supports experience reporting, configuration workflows, documentation search, and reusable workflow prompts.
Available actions depend on your Convert permissions and connection scope.

Hosted Connection

Use a client that supports remote MCP over Streamable HTTP with OAuth.
Add the following server URL in the client's remote connector settings:

https://mcp.ai.app.convert.com/mcp

Complete the Convert sign-in and permission approval flow in your browser.
Choose the account and selected projects needed for your workflow.
Eligible users can explicitly authorize Entire account, including future projects and select a maximum role.
Selecting every current project does not grant access to future projects.
Effective access cannot exceed either the approved role ceiling or your current Convert role.

Do not add API keys, secrets, or bearer tokens to the server URL.
Do not paste credentials into an AI conversation.
The hosted connector uses OAuth, not query-string authentication.

Client configuration differs between products.
Use your client's remote connector setup rather than a local stdio configuration for this hosted connection.

Local Connection

Clients that support local stdio servers can run the Convert MCP npm package.
This connection uses your Convert API key and secret from Account settings > API Keys.
These credentials are not an OAuth client ID and client secret.
Keep them in your local client configuration or its supported secret store.

{
  "mcpServers": {
    "convert": {
      "command": "npx",
      "args": ["-y", "@convertcom/mcp-server@latest"],
      "env": {
        "CONVERT_API_KEY": "YOUR_CONVERT_API_KEY",
        "CONVERT_API_SECRET": "YOUR_CONVERT_API_SECRET",
        "TOOLS_FOR_CLIENT": "reporting"
      }
    }
  }
}

Use the configuration location documented by your MCP client.
The local reporting default and readOnly profile expose semantic reads, including POST-backed reporting endpoints.
The all profile also exposes writes and deletes, subject to Convert permissions and MCP guardrails.
Changing a local profile does not grant additional backend permissions.

Safe Workflows

Start with a read-only request that names the project and date range.
For example: "Summarize the active experiences in my selected project over the last 30 days."
Fetch ordinary tool results before requesting an optional interactive view.
Interactive views are read-only and depend on client support.
Reusable MCP prompts also depend on the client's prompt selection support.

New experiences default to draft when their status is omitted.
Setting experience status to active requires explicit user confirmation metadata with a reason.
Deletion requires an exact user request, a second confirmation, and metadata bound to the exact numeric resource ID.
The client must obtain these human decisions before supplying confirmation metadata.
Metadata records stated intent but does not independently prove a human approved the action.
Confirmation never grants backend authorization.

When Convert requires a governed approval, submit an approval request only after explicit user instruction.
Approving or rejecting a request is a separate decision.
An approval does not automatically execute or retry the original action.
Read back any authorized change to verify its persisted state.

Compatibility

MCP compatibility depends on the client, transport, authentication, schemas, and model.
A working connection does not establish reliable performance for every model.
The package guidance recommends models with strong multi-step reasoning capabilities.
There is no small-model performance claim on this page.

Consult the package setup and configuration reference for version-specific details.