Free MCP Server Config Generator
An MCP server config generator builds the JSON (or TOML) block that tells AI clients like Claude Code, Cursor, or Codex CLI how to launch and connect to Model Context Protocol servers. Pick your client, choose from popular servers (GitHub, Postgres, Slack, Filesystem, and more), plug in your credentials, and copy a validated config block for the right config file.
~/.claude.json (user-scoped).mcp.json in project root (project-scoped)
Claude Code reads project-scoped .mcp.json automatically. Restart any open Claude Code session after adding a server.
{
"mcpServers": {}
}How to set up an MCP server
- Pick your MCP client. Choose Claude Code, Cursor, Codex CLI, Claude Desktop, or VS Code. The generator switches output format and target file path to match.
- Choose an MCP server. Pick from popular servers like Filesystem, GitHub, Postgres, Slack, or Brave Search. Or pick Custom stdio / Custom HTTP to define your own.
- Fill in credentials and paths. Enter the path, connection string, or API token the server needs. Required fields are marked with an asterisk. Secret values are masked by default.
- Copy the config. Click Copy to grab the JSON or TOML block. For Claude Code stdio servers you also get an equivalent `claude mcp add` CLI command.
- Paste and restart. Save the config to the file path shown for your client (e.g. ~/.claude.json or .cursor/mcp.json), then restart the client so it picks up the new server.
Frequently Asked Questions
What is an MCP server?
+
An MCP server is a small program that exposes tools, resources, or prompts to an AI client over the Model Context Protocol. Clients like Claude Code, Cursor, Codex CLI, Claude Desktop, and VS Code launch MCP servers as separate processes (or connect to them over HTTP) and call their tools while reasoning. Common examples are filesystem access, GitHub, Postgres, and Slack servers.
Where do I save the MCP config file for Claude Code?
+
Two options. For user-scope (every project), Claude Code reads ~/.claude.json. For project-scope, drop a .mcp.json file in the repository root and Claude Code picks it up only when running inside that repo. You can also run `claude mcp add <name> -- <command>` from your terminal to write the entry without editing JSON by hand. Restart any open Claude Code session after changes.
How is Cursor's MCP config different from Claude Code's?
+
The JSON shape is identical. Both use a top-level `mcpServers` object keyed by server name. The difference is the file path: Cursor reads ~/.cursor/mcp.json globally and .cursor/mcp.json inside a project, while Claude Code reads ~/.claude.json or .mcp.json. You can copy the same `mcpServers` block between them.
Why does Codex CLI use TOML instead of JSON?
+
Codex CLI's main config file (~/.codex/config.toml) is TOML, so MCP servers go inside it as [mcp_servers.<name>] tables. The fields (`command`, `args`, `env`) are the same as the JSON clients, just in TOML syntax. This generator emits the right format automatically when you pick the Codex CLI client.
How do I add a custom MCP server that isn't in this list?
+
Pick "Custom stdio Server" for a local process or "Custom HTTP / SSE Server" for a remote endpoint. Paste in the command (or URL) you'd run yourself, name the server, and the generator will produce the right JSON or TOML block. You can edit the args array directly in the output if you need extra arguments.
Do I need to restart my client after adding a new MCP server?
+
Yes. Every MCP client reads its config file at startup, then spawns server processes from there. After editing the file, quit and relaunch Claude Code, Cursor, Claude Desktop, or VS Code (or run `Reload Window` in VS Code). Codex CLI re-reads config on the next invocation.
MCP servers add tools. Tornic adds repeatability.
MCP servers give your AI client new capabilities. Tornic turns those capabilities into deterministic, predictable workflows that run the same way every time, without per-call API charges, by reusing your existing Claude Code, Cursor, or Codex CLI subscription.
Try Tornic