Octri

Install

The server ships as @octri/mcp and runs through npx, so there is nothing to install ahead of time. You need your project id, which you'll find in the dashboard.

The hosted server

Octri hosts every project's server, so a client connects to a URL with nothing to install:

text
https://api.octri.dev/api/v1/public/mcp/YOUR_PROJECT_ID/mcp

With sign-in on, the client opens a browser the first time it connects and the person signs in there. MCP Studio shows every snippet below with your project filled in.

Claude Code

bash
claude mcp add --transport http my-api https://api.octri.dev/api/v1/public/mcp/YOUR_PROJECT_ID/mcp

Cursor

In .cursor/mcp.json. With sign-in on, MCP Studio also has an Add to Cursor link that does this in one click.

json
{
  "mcpServers": {
    "my-api": { "url": "https://api.octri.dev/api/v1/public/mcp/YOUR_PROJECT_ID/mcp" }
  }
}

VS Code

In .vscode/mcp.json, or through the Add to VS Code link.

json
{
  "servers": {
    "my-api": { "type": "http", "url": "https://api.octri.dev/api/v1/public/mcp/YOUR_PROJECT_ID/mcp" }
  }
}

Claude Desktop

Open Settings, then Connectors, choose Add custom connector and paste the URL. Claude Desktop connectors cannot send credential headers, so turn sign-in on for operation tools, or install the .mcpb bundle below.

Without sign-in

Send the credential as a header on each request. Octri does not store it:

bash
claude mcp add --transport http my-api https://api.octri.dev/api/v1/public/mcp/YOUR_PROJECT_ID/mcp --header "x-octri-api-token: YOUR_API_TOKEN"

The header carries your credential for the API being called: x-octri-api-token for bearer and OAuth 2, x-octri-api-key (with x-octri-api-key-header if it isn't X-API-Key) for API keys, x-octri-api-username and x-octri-api-password for basic auth.

The hosted server calls only public addresses. To try an API on localhost or a private network, run @octri/mcp locally.

One-click bundle for Claude Desktop

MCP Studio has a download link for a .mcpb file. Opening it installs the server in Claude Desktop, which asks for the credential during the install, so nobody edits a config file. See packages and plugins.

Claude Code plugin

From the Starter plan, MCP Studio can offer your server as a Claude Code plugin. People add your marketplace once:

text
/plugin marketplace add https://api.octri.dev/api/v1/public/mcp/YOUR_PROJECT_ID/claude/marketplace.json

See packages and plugins.

Claude Code, locally

bash
claude mcp add my-api --env OCTRI_API_TOKEN=YOUR_API_TOKEN -- npx -y @octri/mcp --project-id YOUR_PROJECT_ID

Claude Desktop

Add to claude_desktop_config.json. On macOS that lives at ~/Library/Application Support/Claude/claude_desktop_config.json, on Windows at %APPDATA%\Claude\claude_desktop_config.json.

json
{
  "mcpServers": {
    "my-api-docs": {
      "command": "npx",
      "args": ["@octri/mcp", "--project-id", "YOUR_PROJECT_ID"]
    }
  }
}

Restart Claude Desktop after editing the file. The config is only read at startup.

Cursor

Add to .cursor/mcp.json in your project root, or ~/.cursor/mcp.json to enable it everywhere.

json
{
  "mcpServers": {
    "my-api-docs": {
      "command": "npx",
      "args": ["@octri/mcp"],
      "env": {
        "OCTRI_PROJECT_ID": "YOUR_PROJECT_ID"
      }
    }
  }
}

VS Code

Add to .vscode/mcp.json.

json
{
  "servers": {
    "my-api-docs": {
      "command": "npx",
      "args": ["@octri/mcp"],
      "env": {
        "OCTRI_PROJECT_ID": "YOUR_PROJECT_ID"
      }
    }
  }
}

Flag or environment variable

--project-id and OCTRI_PROJECT_ID do the same thing. The flag wins if both are set.

Prefer the env var for a shared config

A .cursor/mcp.json committed to your repo gives every contributor the server without each one editing their own config. The env form keeps that file free of anything machine-specific.

Verifying it works

Before you open a client, check the setup from a terminal:

bash
npx -y @octri/mcp doctor --project-id YOUR_PROJECT_ID

It confirms the project and its tools load, the base URL and credentials are set, and makes one real read call, then says what to fix if anything fails. Add --no-call to skip the call.

Then ask the assistant something only your API can answer. MCP Studio lists every call agents make, as they happen, under Try it live: the tool, how it ended, which client made it and how long it took. A call appearing there is the server working end to end.

If nothing shows up, see Troubleshooting.

Publishing under your own name

MCP Studio can publish this server to npm or PyPI as your own package. See packages and plugins.

Next