How to Connect Grok Build to Skill Store 33's MCP Server

Short answer: in a terminal, run grok mcp add --transport http skill-store https://mcp.ss33.ai, then start Grok Build and ask it to find a skill for your task. That's all you need. No sign-in or key is required. A free Skill Store 33 account adds a key that raises your search limit, and you pass it with --header.

Grok Build is xAI's coding agent. You run it as grok in a project folder. It supports MCP servers, local or remote, and stores them in a TOML config file. Everything below was checked against xAI's Grok Build docs on Oct 30, 2026.

What you are connecting

Skill Store 33 runs a public MCP server at https://mcp.ss33.ai. It speaks Streamable HTTP, so there is nothing to install or run on your machine. It gives your agent three tools:

search_skills(query, area): describe a task and get back a few matching skills, with what each one does.

get_skill(skill_ref): fetch the chosen skill's SKILL.md so the agent can follow it.

report_skill_outcome(skill_ref, worked, notes): tell the store whether the skill helped.

Without an account, search_skills allows 5 searches a minute per address. A free account gives you a key with 60 searches a minute. The other two tools are not rate limited.

Step 1: Install Grok Build

If you don't have it yet, xAI's install line for macOS and Linux is:

curl -fsSL https://x.ai/cli/install.sh | bash

On Windows, in PowerShell:

irm https://x.ai/cli/install.ps1 | iex

Then cd into a project and run grok. On first launch it opens a browser to sign you in. Where there is no browser, xAI's docs say to set the XAI_API_KEY environment variable instead.

Step 2: Add the server with the CLI

Grok Build's docs show grok mcp add with --transport http for remote servers. For Skill Store 33:

grok mcp add --transport http skill-store https://mcp.ss33.ai

Grok Build keeps your user settings in ~/.grok/config.toml (on Windows, %USERPROFILE%\.grok\config.toml). To add the server for one repository only, add --scope project. That writes .grok/config.toml in the current folder, which you can commit so your team gets the same server.

Or add it in config.toml

If you'd rather edit the file, each server is a table under mcp_servers. A remote server needs only a url:

[mcp_servers.skill-store]

url = "https://mcp.ss33.ai"

The same table also accepts headers, startup_timeout_sec (default 30) and tool_timeout_sec.

Step 3: Add your account key (optional)

Five searches a minute is fine for trying it out. For regular use, create a free account at https://ss33.ai/skills/account, then click Create key. Send the key as an Authorization header. Grok Build's --header flag does this, and you can repeat it for more headers:

grok mcp add --transport http skill-store https://mcp.ss33.ai --header "Authorization: Bearer YOUR_KEY"

If you already added the server without a key, remove it first with grok mcp remove skill-store, then add it again with the header.

To keep the key out of the file, put it in an environment variable. Grok Build expands ${VAR} in url, command, args, env and headers. In config.toml:

[mcp_servers.skill-store]

url = "https://mcp.ss33.ai"

headers = { "Authorization" = "Bearer ${SS33_KEY}" }

Then export SS33_KEY in your shell before starting grok. SS33_KEY is just a name we picked. Use any variable name you like. Don't commit a project config that contains the key itself.

Step 4: Check that it works

From the shell, grok mcp list shows your configured servers, and grok mcp doctor skill-store checks the config and the connection. Both take --json if you want output a script can read.

Inside a session, /mcps opens the MCP tab. You should see skill-store there. Space turns a server on or off, and r refreshes the list.

Then give it a real task. For example:

Find a skill for writing a GitHub Actions workflow that runs pytest on every pull request, then use it.

You should see Grok Build call search_skills, pick a result and call get_skill to read it. If it answers from memory instead, ask it directly to use the skill-store search_skills tool.

Troubleshooting

Server missing or not connecting: run grok mcp doctor first. It checks both your configuration and connectivity.

Server listed twice, or one you didn't add in Grok: Grok Build also loads MCP servers from ~/.claude.json, .cursor/mcp.json and project .mcp.json files. Those have lower priority than config.toml. If you already added skill-store in Claude Code or Cursor, Grok Build may pick it up from there.

Searches start failing after a burst: you have probably hit the 5-a-minute limit for anonymous use. Wait a minute, or add an account key as in Step 3.

FAQ

Do I need an account to use Skill Store 33 in Grok Build?

No. The server needs no sign-in. Without an account you get 5 searches a minute. A free account key raises that to 60.

Can I share the server with my team?

Yes. Add it with --scope project and commit .grok/config.toml. Each person then adds their own key through an environment variable, not in the committed file.

Related guide: One MCP server, 13 coding agents: the config keys that differ

Let your agent search Skill Store 33

Sources

xAI, Grok Build docs: MCP Servers, https://docs.x.ai/build/features/mcp-servers (read Oct 30, 2026)

xAI, Grok Build docs: Settings, https://docs.x.ai/build/settings (read Oct 30, 2026)

xAI, Grok Build docs: Overview, https://docs.x.ai/build/overview (read Oct 30, 2026)

This article was created with the help of AI (AI-Assisted). Every config key and command was checked against the official docs in the Sources list. ReadyAI is accountable for the content.

Keep reading