Connect JetBrains AI Assistant to Semogram

Use JetBrains AI Assistant to inspect your Semogram workspace from the conversation where you already work. For example, compare a delivery summary with its underlying orders and receipts, or inspect a published workflow before deciding to run it.

The steps below follow the client’s documented MCP setup. This client connection has not yet been verified end to end against Semogram; complete the first-read checks below to verify your workspace access.

Install

Before you start

You need a Semogram account with access to a workspace, an installed client with MCP tools enabled, and a model that can call tools. Your organization may control which external servers and assistant providers you can use.

In Semogram, open workspace Settings → Connect an AI assistant and copy the workspace MCP URL. Its form is:

https://platform.semogram.com/api/mcp/workspaces/<WORKSPACE_ID>

Use the copied URL wherever the example says <YOUR_WORKSPACE_MCP_URL>. A workspace name cannot replace its UUID. The URL identifies the workspace and does not grant access by itself. Connecting an assistant does not import records or create a project; connect your data if there is nothing to inspect yet.

Add the connection and sign in

This guide uses mcp-remote as a local OAuth bridge. JetBrains documents remote HTTP connections, but its AI Assistant setup documentation does not establish a Semogram-compatible interactive OAuth flow. The bridge handles that sign-in while AI Assistant communicates with a local STDIO server.

You need Node.js and npx available to the IDE, plus permission to run this third-party adapter. Review mcp-remote and use your organization’s approved version.

Open Settings → Tools → AI Assistant → Model Context Protocol (MCP). Add a server using the STDIO connection option and this JSON configuration:

{
  "mcpServers": {
    "semogram": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "<YOUR_WORKSPACE_MCP_URL>",
        "--transport",
        "http-only"
      ]
    }
  }
}

Replace the URL and, for managed installations, replace mcp-remote with your approved pinned package version. Choose global or project scope, save, and apply. The adapter opens a browser for Semogram sign-in and consent. Complete that flow on the same machine, then inspect the server status in AI Assistant.

This config belongs under AI Assistant’s client settings, not the IDE’s built-in MCP Server settings, which expose IDE tools to other assistants.

Make your first read

In AI Assistant chat, start with a read-only request:

Use Semogram to discover the available capabilities, confirm the workspace name and ID, and list the projects I can access. Do not change anything.

The assistant should call discover_capabilities, workspace_get, and project_list. Compare the returned workspace with the one you intended and choose a project. Workspace-scoped resources do not take a projectId; project operations on this workspace connection do.

Then use a record or query that actually exists in that project:

Inspect our published delivery summary query. Explain its inputs, show its latest run, and identify the order and receipt records behind the result. If it does not exist, say so rather than creating a replacement.

Replace the example query name with yours. Inspect the tool calls and returned references, rather than accepting a conversational answer as proof of access. A tool list confirms discovery, not that your data is complete or the answer is correct. Read a run to inspect the result.

Manage the connection

Inspect the Status column under AI Assistant’s MCP settings. Disable or remove the Semogram entry to stop the local bridge. This does not revoke cached bridge authentication or delete workspace records; use the adapter’s documented token cache controls when you also need to clear saved authentication.

Removing a client configuration does not delete Semogram data. Each additional workspace should have its own clearly named connection and its own copied URL.

Bridge troubleshooting

If the IDE reports that npx cannot be found, use its full executable path from your Node installation. If sign-in opens but never returns, check that the browser can reach the adapter’s loopback callback on this machine. Remote/headless IDE sessions need a separately planned authentication setup; a desktop callback is not automatically reachable from a container.

FAQ

What permissions does the assistant have?

An OAuth connection uses the signed-in account’s current workspace permissions. A missing project or tool can reflect those permissions. An assistant approval prompt does not replace Semogram’s server-side operation approvals.

Ask to inspect drafts before executing workflows or changing records. Keep the returned job or operation ID when work runs asynchronously and check its final status. Data returned to the client is handled by its assistant provider; use your organization’s approved account and data-sharing policy.

Source-write tools have additional API-key requirements described in the MCP reference. They are not enabled simply by reconnecting an OAuth client.

Why is the connection failing?

For client-specific failures, inspect AI Assistant’s server status and adapter logs. Before sharing logs, remove tokens, private records, and personal information. See MCP troubleshooting for server error details.

Where can I learn more?

Return to choose your assistant, review tools and scopes, or consult the official JetBrains AI Assistant MCP documentation for controls that vary by version.