Connection
| Available on request for corporate clients. |
You need an account to work Engee, agent-based client with support MCP and a Large Language Model (LLM) capable of invoking tools. The cloud client needs access to the server Engee from its own infrastructure; to the local one — from the computer where the client is working.
|
Client interface. LLM clients are constantly evolving and changing, so the names of sections, buttons, and fields in this manual may differ from the current interface. For more information about integration, please contact technical support Engee; according to the settings of the client itself — in support of its developer. |
To access the work environment, send a Personal Access Token (PAT) to the chat with the agent. The PAT release procedure is described in external API documentation Engee.
Connection parameters
| Field | Meaning |
|---|---|
Display Name |
|
URL of the MCP server |
Write to engee@engee.com to specify the access mode for your organization |
Transport |
|
Authorization when connecting the MCP |
|
Access to the user’s work environment |
PAT Engee in the parameter |
Authorization of the MCP connection and access to the user environment are different steps. The client can receive a list of tools without PAT. For operations with the work environment, the agent passes PAT in the arguments of the tools. In these configurations, you do not need to add PAT to the URL or header. Authorization.
Creating a PAT
-
Save the working files and stop the session Engee before the token is issued. A PAT created during a running session will become effective only after it is stopped and restarted.
-
Open personal account Engee and go to Communication and Security → Personal access tokens.
-
Click Add a new token, set the name, expiration date, and permission To access the API Engee.
-
Click Add Personal Access Token and copy the value in full, including special characters. You will not be able to view it again.
-
Launch Engee after creating the token, either assign the launch to the agent.
For the first check, send the PAT to the chat with the agent. Do not put PAT in public files, shared configurations, or published reports. You can revoke it in the tokens section of your personal account. Engee.
Customer’s choice
| Client | Setting up method |
|---|---|
ChatGPT, Codex |
Application in ChatGPT; MCP settings or TOML in Codex |
Claude.ai, Claude Code |
User connector; CLI or |
Grok |
Custom MCP Connector |
Yandex AI Studio |
External MCP server in the MCP Hub |
Giga Cowork |
External connector parameters; specific delivery features |
OpenCode |
|
llama.cpp |
MCP in llama-ui or stdio-bridge for llama-server |
Ollama |
Local model and MCP in OpenCode |
LM Studio (Bionic) |
|
The first session
After setting up, select Engee MIND in the current chat and send:
Используй Engee MIND. Вызови engee_status, покажи состояние среды и доступные навыки. Пока не изменяй модели и файлы.
Without PAT, authorization instructions are expected, not the state of your environment. The actual status is expected after the PAT transfer. To start a stopped environment, send:
Запусти мою рабочую среду Engee, если она остановлена, и загрузи навык engee_model. Сообщи, когда среда будет готова.
It usually takes tens of seconds to start; a cold start with library preparation may take longer. Verification is considered successful when real tool calls are visible, the environment status is received, and the skill is loaded. The agent’s text response without calling the tools is not a connection check.
Typical problems
| Problem | What to check |
|---|---|
|
URL and transport: we need Streamable HTTP, not the outdated HTTP with SSE |
There are tools, operations with Engee rejected |
PAT, resolution «API Access Engee», the completeness of the meaning, the presence of |
The agent only describes the actions |
Is the server connected to the current chat, are the tools allowed, does the LLM tool support calling |
The call is interrupted when the environment starts |
Tool execution timeout in the client; it differs from tool discovery timeout |
No model found after downtime |
The environment could have stopped; start it and load the previously saved one. |
The file was created, but did not appear in the chat |
The agent must call |
The configuration format depends on the client: there is no universal JSON for all applications. When adding to an existing configuration, merge the relevant sections, saving the other servers and settings.