Use tolz.me with AI agents
Every tool on Tolz.me is also a Model Context Protocol tool. Point an agent at one URL and it can build Discord timestamps, format text, convert cron expressions, work out chmod values, sort lists, and search the guides.
The same logic runs in both places. The tool you click on the site and the tool an assistant calls use the same functions, so the answer you check by hand is the answer the agent got. There is no account to create, no key to request, and nothing to install on the client side beyond whatever your agent already uses.
The MCP server
The server URL is https://tolz.me/api/mcp. It speaks Streamable HTTP and needs no authentication. The tools are read-only and deterministic, so the same input always returns the same output and nothing on the server changes.
Two of the tools help an agent find its way:
list_toolsreturns every tool page with its category, a one-line description, and the URL. Pass a category to narrow the list.search_blogsearches the guides by title, description, and tag.
Example prompts:
- "List the Tolz.me tools in the developer category and give me the URLs."
- "Search the Tolz.me guides for cron expressions."
- "Convert the cron expression 0 9 * * 1-5 into plain English."
What the server exposes
The server registers fifteen tools. Thirteen of them do the work itself, and two of them help an agent find the rest of the site. Each one takes plain arguments and returns an answer.
Discord
Formatting and text tools that follow the rules Discord actually applies.
discord_timestampBuilds a timestamp tag such as <t:1715777220:f>. Give it a date and time or a Unix time, and leave the style out to get all seven styles at once.discord_text_formatWraps a message in bold, italic, underline, strikethrough, spoiler, inline code, or a code block.fancy_font_textRestyles text in one of the Unicode fonts the site uses, for a display name, a bio, or a message.small_textReturns the subtext prefix, small caps, superscript, and subscript versions of a message.spoiler_textHides a message behind spoiler bars, either as one block, per word, or per letter.
Text
Cleanup for text that is about to be pasted somewhere else.
sort_linesSorts a list, optionally removing duplicates, empty lines, and stray spaces.remove_em_dashesReplaces em dashes, and optionally en dashes and double hyphens, with a comma, semicolon, colon, hyphen, or your own text.remove_invisible_charactersFinds and removes hidden zero width characters and unusual spaces, while keeping emoji sequences together.
Developer
The everyday conversions that get typed out by hand.
chmod_calculateReads permissions written as 755, rwxr-xr-x, or -rwxr-xr-x and returns the octal, symbolic, and ls forms, the commands to run, and a plain English reading.cron_translateTranslates a cron expression into plain English and lists the next run times.unix_timestamp_convertTurns a Unix timestamp into a readable date, or a date and time into a Unix timestamp.
Calculator
Arithmetic that is easy to get wrong when you are working it out in your head.
hours_between_timesTotals shift hours across several clock times, subtracts unpaid breaks, and estimates gross pay from an hourly rate.aspect_ratioReduces a pixel size to its simplest ratio, checks a decimal ratio, or works out one side of a target ratio from the other.
Finding things
Two tools that let an agent explore the site before it starts working.
list_toolsReturns every tool page with its name, category, one-line description, and URL. Pass a category to narrow the list.search_blogSearches the guides by title, description, and tag.
What a tool call returns
Every tool answers with two things. The first is plain text that reads well on its own, which is what most assistants show you. The second is a structured payload with the same information in fields, so a program can read a value without parsing a sentence.
When an input cannot be handled, the tool says so in a readable way and marks the result as an error, rather than returning something broken. An unreadable cron expression comes back as a sentence explaining what was wrong with it, so the agent can pass on a sensible message instead of an empty answer.
Arguments are described in the schema, so a client can show what a tool expects before it calls it. Most arguments have sensible defaults: leave the timestamp style out and you get all seven styles, leave the count out of a cron lookup and you get ten run times, leave the replacement out of the em dash tool and it uses a comma.
Calling the server yourself
Most people never need this, because a client library handles the handshake for you. If you are writing your own integration, the endpoint is a standard JSON-RPC 2.0 POST. A tool call looks like this:
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "cron_translate",
"arguments": {
"expression": "0 9 * * 1-5",
"count": 3
}
}
}The same request with curl:
curl https://tolz.me/api/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"cron_translate","arguments":{"expression":"0 9 * * 1-5"}}}'If you drive it by hand, send an initialize request first and keep the session id that comes back in the response header, because the transport ties later calls to that session. An MCP client library does this for you. To see the schema of every tool without calling one, ask for the tool list instead.
ChatGPT (web)
- Open Plugins in ChatGPT and choose +.
- Choose Add custom MCP server.
- Name:
tolz.me - Server URL:
https://tolz.me/api/mcp - Authentication: No authentication
- Tick I understand and want to continue.
- Choose Create as a plugin.
- Type
@in the prompt and select it.
Availability depends on your plan and workspace settings. OpenAI documents the steps in its help article on custom MCP servers.
Claude (claude.ai or the desktop app)
- Open Customize, then Connectors, then Add.
- Choose Custom, then Web.
- Name:
tolz.me - URL:
https://tolz.me/api/mcp - Choose Continue.
- Authentication: No sign in
- Choose Add.
The free plan allows one custom connector. Team and Enterprise owners add it in Organization settings, under Connectors. Anthropic documents the steps in its help article on custom connectors.
Or open the one-click connector link , which fills in the name and URL for you.
Cursor
Add the server to ~/.cursor/mcp.json for every project, or to .cursor/mcp.json in a single project folder:
{"mcpServers":{"tolz":{"url":"https://tolz.me/api/mcp"}}}Browser agents
Tool pages register WebMCP tools through document.modelContext, so a browser agent can work the tool on the page itself. In Chrome, WebMCP is available through the origin trial for versions 149 to 156, or you can turn on chrome://flags/#enable-webmcp-testing.
A tool page registers the tools that suit it, and some register more than one. The Discord timestamp page offers generate_discord_timestamp to build a tag and list_discord_timestamp_styles to list the seven display styles. The mention formatter registers format_discord_mention, the embed generator registers build_discord_embed, and the text cleanup pages register their own versions of the sorting, em dash, and invisible character tools. The result is written back into the visible form, so you can see what the agent did and adjust it before you use it.
Every page also offers two site-wide tools, list_tolz_tools and search_tolz_tools, so an agent can look up the rest of the catalogue from any page.
Read the site without MCP
- /llms.txt is a short list of every tool and guide.
- /llms-full.txt holds the full text of every tool page and guide.
- /.well-known/mcp-server-card is a JSON card with the server URL and transport.
- /.well-known/ai-tools.json is a machine-readable catalogue of every tool, with keywords and the categories they sit in.
The two text files are useful when you want the wording of a tool page or a guide in one place, without calling a tool at all. The catalogue is useful when you are building an index of your own.
Frequently asked questions
Do I need an API key?+
No. The server needs no authentication and there is no key to request or rotate. Point a client at the URL and it can list the tools and call them.
Can the tools change anything on my account?+
No. Every tool is read-only and deterministic. The same input always returns the same output, nothing is stored, and no tool posts a message, changes a role, or touches a Discord account.
Can I call it from a script instead of an assistant?+
Yes. The endpoint speaks the standard Model Context Protocol over Streamable HTTP, so any MCP client library can talk to it. A raw call is a JSON-RPC 2.0 POST, and the examples below show the shape of the request.
What is the difference between the MCP server and WebMCP?+
The MCP server is a remote endpoint that an assistant calls over the network. WebMCP is the browser-side version: a tool page registers its own tools with the browser through document.modelContext, so an agent that is already looking at the page can work the tool directly.
Why does the MCP server expose fewer tools than the site has?+
The site has tools that need a file, a preview, or a decision from a person, and those do not translate into a single call and answer. The MCP server keeps the tools where the whole job can be done from the arguments alone.