tuqo
IDE · tqk_ key in a header

Tuqo in ZCode (GLM)

ZCode is Z.ai's desktop agent for GLM models. MCP servers are added in the app settings: a remote server takes the HTTP type, an address and headers. User-level servers live in ~/.zcode/cli/config.json, project-level ones in .zcode/config.json. Tuqo connects with a tqk_ project key in the Authorization header.

Project key in the Authorization header · scope of your choice
1

Create a project key

In the Tuqo panel at app.tuqo.dev → project → AI & API → create. Format tqk_…, shown once.

2

Open MCP settings

ZCode → Settings → MCP Servers → New MCP Server. Scope: User (all projects) or Workspace (this folder only).

3

Describe the server

Name tuqo, type HTTP, the MCP address; under Headers (optional), Authorization with the value Bearer tqk_… (example below). Click Add.

4

Deploy

Ask the agent to create a site and publish it. It calls the Tuqo tools and returns the live link.

Name:    tuqo
Type:    HTTP
URL:     https://mcp.tuqo.ru/mcp
Headers: Authorization = Bearer tqk_xxx_yyy

The Tuqo panel is in English, and the labels here are the ones it shows.

ZCode (GLM) + Tuqo: frequently asked

I use GLM through Claude Code or OpenCode +

Then connect Tuqo in the client you work in: MCP is set up in the agent, not in the model. See the Claude Code and OpenCode pages under /connect/.

Can I bring the server over from Claude Code? +

The MCP Servers page has an import that reads ~/.claude/settings.json and the Codex and OpenCode configs. A server added with claude mcp add lives in another file (~/.claude.json) and may not be found; then add Tuqo by hand. After importing, check the Authorization header on the tuqo entry.

User or Workspace? +

User makes the server available in every project on this machine; Workspace only in the current folder (the .zcode/config.json file). Better not to put an entry with a key in the project file: it's easy to commit.

How many tools does the agent get? +

All 52 Tuqo tools: sites, deploys, domains, rollbacks, gated access, form submissions. If the model's context is narrow, use a read-only key and a full one for deploys.

Connect another AI