Custom ToolsBeta

Built-in commands cover the web, your files, and your own history. A custom tool connects Glaido to anything else with an API, like your task manager, your notes, or your company's internal systems.

You speak as usual, Glaido calls the tool, and the answer is ready to paste at your cursor.

Add your first tool

You don't build tools by hand, and you don't need to know how to code. An AI coding agent such as Claude Code, Cursor, or Codex builds them for you. Your first one takes about two minutes:

1. Open your coding agent.

2. Paste this prompt and let it run:

   I'm setting up custom tools for the Glaido dictation app. Read this guide and
   follow the instructions for coding agents: https://docs.glaido.com/docs/beta/tools
   
   Build the hello-world weather example from the guide (current weather for any
   city, using the free Open-Meteo API, no API key needed). It's a fixed demo:
   just run the commands from the guide without planning or questions, then tell
   me the exact folder to pick when I click Import folder on Glaido's Commands page.

3. Import the result. When the agent finishes, open Commands in the Glaido sidebar, go to the Custom tab, click Import folder, and select the folder it created. Enable the server with its toggle.

Hold your commands hotkey and ask "What's the weather in Amsterdam?"

The example needs no API key and no account, so nothing can go wrong with credentials. The agent installs anything else it needs along the way.

No coding agent yet?

Install one first; it's a normal desktop app plus a terminal command. Claude Code is a good default. Once it's installed and signed in, paste the prompt above into it.

Build your own

Once the weather example works, use the same prompt, but replace the middle paragraph with the tool you want. For example:

• "Connect Glaido to my Notion workspace"
• "Make Glaido able to send Slack messages"
• "Give Glaido access to my company's internal API"

The agent decides whether a ready-made server already exists for your service or builds a small custom one, wires up any API keys in a local file, and hands you a folder to import. It will tell you which keys to set and where to get them.

Start from an empty server instead

If you'd rather have the folder ready before you open your coding agent, Glaido can scaffold one for you. On the Custom tab, click New server.

Glaido watches that folder, so the tools show up on the Commands screen as your agent saves them. You don't have to import it afterwards.

Install the skill

The prompt above makes your agent fetch the instructions fresh each time. If you plan to build more tools, install the skill from the glaido-skills repository instead. Your agent then knows the whole workflow by heart and triggers it automatically on any request like "make Glaido able to send Slack messages", no prompt needed. You can also ask the agent questions about the skill itself to understand how Glaido tools work.

git clone https://github.com/daveebbelaar/glaido-skills.git
mkdir -p ~/.claude/skills
cp -r glaido-skills/creating-glaido-mcp-servers ~/.claude/skills/

git clone https://github.com/daveebbelaar/glaido-skills.git
mkdir -p ~/.agents/skills
cp -r glaido-skills/creating-glaido-mcp-servers ~/.agents/skills/

These commands are safe to re-run: mkdir -p does nothing if the folder already exists, and copying only adds or updates this one skill without touching any other skills you have installed. On Windows, run them in PowerShell.

Where your tools live

Two folders come out of the box, and it's worth settling on one. New server scaffolds into ~/Glaido/servers. The coding-agent prompt at the top of this page puts the weather example in ~/Glaido/glaido-mcp-servers. Pick the one you want to live in, tell your agent to build there, and move anything that landed in the other.

Keep every new server in that one folder. Once you have a few, turn it into a git repository and push it to GitHub. You get version history for your tools, and setting up a new machine becomes clone and import. Your coding agent can do this for you. You can keep servers anywhere you like, but if you move one, import it again from its new location.

Deciding what runs without asking

Approval covers the servers you added, plus one built-in command. Analyze image asks before it uses a picture from your clipboard. See Approving a clipboard picture.

Every other built-in command runs whenever the model calls it, and its switch on the Built in tab decides whether the model sees it at all. That includes Files and apps, which reads and opens things on your disk without asking first. Turn it off there if you'd rather it didn't.

For your own servers, every tool has an approval policy. Click a server card on the Commands page and set it per tool:

Tools that only read data default to Auto; everything else defaults to Ask. When a tool set to Ask is about to run, an approval card appears in the command window:

Allow once runs the tool this one time, Always allow switches it to Auto, and Deny declines the call. While the card is visible, pressing Enter chooses Allow once and never pastes.

The card doesn't wait forever. Leave it for 60 seconds and nothing runs. Glaido says it stopped waiting and asks whether to try again.

Why local, specific tools?

Glaido uses a fast model so answers come back in about a second, and fast models work best with a small set of specific tools.

That is why Glaido runs tools as small servers on your own machine, rather than connecting to large cloud MCP servers. Those typically expose dozens of broad, general-purpose tools designed for long-running agents like Claude Code, and loading them all makes a fast model slower and less accurate. A tool that does one job well ("get the weather", "add a task", "search my notes") keeps Glaido fast.

Reference

Custom tools are Model Context Protocol (MCP) servers that run as local processes on your machine. Manage them from the Custom tab:

• New server scaffolds an empty one for you, as described above.

• Import folder adds a folder that contains an mcp.json (or .mcp.json) file. This is what the coding agent flow produces.

• Edit, under Config file at the bottom of the tab, opens Glaido's MCP config file, ~/Glaido/mcp.json, in your editor. It uses the standard mcpServers format shared by most MCP clients:

  {
    "mcpServers": {
      "my-server": {
        "command": "npx",
        "args": ["-y", "@example/mcp-server"],
        "env": { "API_KEY": "your-key" }
      }
    }
  }

Glaido reloads the server list when the file changes.

A server card shows an extension icon, the server's name, a one-line description, and up to three of its tools as chips, with a +N more chip for the rest. On the right is the switch that enables it. Nothing else appears there unless the server failed to start, which adds a warning badge to the icon.

Click the card for its details: the status dot, every tool with its approval policy, the failure message if there is one, the configuration Glaido runs, and buttons to refresh the tools, open the folder, or delete the server.

Glaido runs each imported server from the folder you imported, so an mcp.json doesn't need any absolute paths.

When a server won't start

A card that failed carries a warning badge. Open it and Glaido names the cause instead of making you read a log:

Where a server stopped with an exit code, Glaido names the code in the message.

Glaido detects the first three by checking your setup, so there's no server output to show. For the rest, whatever the server printed appears under the message in the details dialog. Ask your coding agent to run the validation script from the skill; it catches all of them before you import.

For coding agents

If you are a coding agent sent here to set up a tool for Glaido:

1. Fetch and read this skill, then follow it exactly: https://raw.githubusercontent.com/daveebbelaar/glaido-skills/main/creating-glaido-mcp-servers/SKILL.md

2. If the user asked for the hello-world weather example, you do not need to read the skill or anything else: it is a fixed demo. Do not plan, do not generate or modify code, do not propose folder locations, and do not ask for approval. Check that uv is installed (install it from docs.astral.sh/uv if not), then run exactly:

   TMP=$(mktemp -d)
   git clone --depth 1 https://github.com/daveebbelaar/glaido-skills.git "$TMP/glaido-skills"
   mkdir -p ~/Glaido/glaido-mcp-servers
   cp -r "$TMP/glaido-skills/examples/hello-world-weather" ~/Glaido/glaido-mcp-servers/hello-world-weather
   python3 "$TMP/glaido-skills/creating-glaido-mcp-servers/scripts/validate_glaido_mcp.py" ~/Glaido/glaido-mcp-servers/hello-world-weather

   (Windows: adapt to PowerShell; clone into $env:TEMP instead of mktemp.) When the validator reports "ready to import", you're done. It already boot-tests the server, so skip any further verification. The mcp.json is intentionally path-free; nothing needs editing. Go straight to the import steps below.

3. For anything else, the skill covers reusing existing third-party MCP servers, scaffolding custom ones, environment setup, the mcp.json schema, validation, and hand-off steps.

4. Whatever you build, your final message must end with the exact import steps for the user: open the Glaido app, click Commands in the sidebar, go to the Custom tab, click Import folder, then in the file browser that opens select the whole folder you created (the folder itself, not a file inside it) and confirm, and switch the server's toggle on. Then suggest one voice command that uses the new tool.

The skill is the source of truth and takes precedence over the summary on this page.