EX MCP Server

# Make EX MCP global in VS Code (Augment) This guide registers the EX MCP server once (User Settings), so every project can use it without copying settings into each repo. ## Prerequisites - VS Code with the Augment extension installed/updated - This repo cloned locally (path in examples below) - API keys (optional but recommended): KIMI_API_KEY and/or GLM_API_KEY; optionally OPENROUTER_API_KEY or CUSTOM_API_URL ## 1) Open VS Code User Settings (JSON) - Command Palette → “Preferences: Open User Settings (JSON)” - You’ll edit your global settings.json (not the workspace .vscode/settings.json) ## 2) Add the global EX MCP server entry Pick ONE of the two options. Option A — use the console script if installed (ex-mcp-server on PATH): ``` { "chat.mcp.discovery.enabled": true, "chat.mcp.autostart": "whenUsed", // or "always" "chat.mcp.servers": { "ex-mcp": { "transport": "stdio", "command": "ex-mcp-server", "args": [], "env": { "DEFAULT_MODEL": "glm-4.5-flash" } } } } ``` Option B — call Python directly (no console script): ``` { "chat.mcp.discovery.enabled": true, "chat.mcp.autostart": "whenUsed", "chat.mcp.servers": { "ex-mcp": { "transport": "stdio", "command": "python", "args": [ "c:/Project/EX-AI-MCP-Server/server.py" ], "env": { "DEFAULT_MODEL": "glm-4.5-flash" } } } } ``` Notes - Keep only one ex-mcp block. If you already have a chat.mcp.servers object, merge entries. - You can switch DEFAULT_MODEL to "auto" to let the server choose per tool. ## 3) Set provider environment variables (Windows) Open PowerShell and set any you plan to use, then restart VS Code: ``` setx KIMI_API_KEY "<your_key>" setx GLM_API_KEY "<your_key>" :: Optional setx OPENROUTER_API_KEY "<your_key>" setx CUSTOM_API_URL "http://localhost:11434" setx CUSTOM_MODEL_NAME "llama3.2" ``` Optional visibility helper (we already enabled a server fallback, but you can set this too): ``` setx INLINE_PROGRESS_IN_TEXT true ``` ## 4) Restart VS Code and verify - View → Output → pick “Augment” or “Augment Agent” - You should see lines like: “MCP Client Connected: ex-server” and tool discovery - In chat, run: listmodels_exai — you should see the models list with a PROGRESS block at the top ## 5) Use in any project Once the User Settings entry exists, every workspace can use EX MCP without any per-project config. ### If a workspace has its own .vscode/settings.json - It can override or add servers, but it is not required - Remove duplicated ex-mcp definitions in repos to avoid confusion; rely on the User Settings one ## 6) Optional: Autostart - "chat.mcp.autostart": "whenUsed" starts EX MCP when a message triggers it - Set to "always" to start it on session open (slightly faster first call) ## 7) Troubleshooting - No dropdown? The UI may not render metadata yet. We inline the PROGRESS block into the primary text, so you should still see steps. - No models found? Run listmodels_exai and check the Output panel for which API keys are detected. - Command not found? Use Option B (python + server.py path) or ensure ex-mcp-server is on PATH. - Windows quoting: use forward slashes or double-escape backslashes in JSON if needed. ## 8) Per-workspace override (only if you want it) Create .vscode/settings.json inside a repo with the same server block to pin a different DEFAULT_MODEL or env just for that project. Otherwise, rely on the global entry. ## 9) What “global” gives you - One place to maintain MCP servers and defaults - All projects share the same EX MCP configuration and progress visibility - You can still override per-project when needed