Connecting a client

Every client needs the same thing: the URL https://mcpdocs.byjg.com/mcp, over streamable HTTP. The documentation is public, so the ByJG server asks for no credentials.

Each section below has two tabs: No token for mcpdocs.byjg.com, and With token for a server of your own that runs with BYJG_DOCS_AUTH_TYPE=bearer. Picking a tab switches every section on the page. In the "With token" tabs, replace <TOKEN> with the token of that server.

The examples name the server byjg-docs; any name works.

Every client here is configured by hand. The server is published in the official MCP Registry as com.byjg/docs, but no editor installs it from there -- that entry is for discovery, see The MCP Registry entry.

Claude Code
Claude Desktop
Codex CLI
Gemini CLI
Qwen Code
Kilo CLI
OpenCode
Cursor
VS Code
JetBrains IDEs
The tools
Troubleshooting

Claude Code

No token
With token

claude mcp add --transport http --scope user byjg-docs \
  https://mcpdocs.byjg.com/mcp

claude mcp add --transport http --scope user byjg-docs \
  https://your-server.example.com/mcp \
  --header "Authorization: Bearer <TOKEN>"

Use --scope user. The default scope is local, which registers the server only for the directory you ran the command in -- for a documentation server you want everywhere, that is almost never what you meant.

Verify:

claude mcp list
# byjg-docs: https://mcpdocs.byjg.com/mcp (HTTP) - ✔ Connected

Remove: claude mcp remove byjg-docs --scope user.

Claude Desktop

Claude Desktop takes a remote server's URL directly, as a custom connector. The same connector works on claude.ai and the mobile apps, and it shows up in Claude Code when you are signed in to the same account.

No token
With token

Open Customize > Connectors, click +, then Add custom connector, and paste the URL:

https://mcpdocs.byjg.com/mcp

Click Add. On a Team or Enterprise plan an owner adds it once under Organization settings > Connectors > Add > Custom > Web; everyone else then enables it from Customize > Connectors.

To use it in a conversation, click + in the chat, open Connectors and toggle it on. A Free plan is limited to one custom connector; paid plans are not.

Claude connects to the server from Anthropic's cloud, not from your machine -- which is why mcpdocs.byjg.com works and a server on your laptop or VPN does not.

Custom connectors only offer OAuth credentials (client ID and secret) under Advanced settings -- there is no field for a static Authorization header, and Anthropic's cloud cannot reach a server on a private network. For a server of your own running with BYJG_DOCS_AUTH_TYPE=bearer, bridge it locally with mcp-remote, a small process that forwards the requests. It needs Node.js 18 or later.

Open Settings > Developer > Edit Config, which opens claude_desktop_config.json, and add:

{
  "mcpServers": {
    "byjg-docs": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://your-server.example.com/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer <TOKEN>"
      }
    }
  }
}

Write "Authorization:${AUTH_HEADER}" exactly like that -- colon, no spaces. On Windows, Claude Desktop does not escape spaces inside args, so the space in Bearer <TOKEN> has to live in the env value instead.

Restart Claude Desktop completely, then check Settings > Developer for the server status.

Codex CLI

Add to ~/.codex/config.toml:

No token
With token

[mcp_servers.byjg-docs]
url = "https://mcpdocs.byjg.com/mcp"

[mcp_servers.byjg-docs]
url = "https://your-server.example.com/mcp"
http_headers = { "Authorization" = "Bearer <TOKEN>" }

To keep the token out of the file, read it from an environment variable instead:

export BYJG_DOCS_TOKEN=<TOKEN>
codex mcp add byjg-docs --url https://your-server.example.com/mcp \
  --bearer-token-env-var BYJG_DOCS_TOKEN

Verify: codex mcp list.

Gemini CLI

No token
With token

gemini mcp add --transport http --scope user \
  byjg-docs https://mcpdocs.byjg.com/mcp

Or edit ~/.gemini/settings.json directly. Note the key is httpUrl; url means the older SSE transport in Gemini CLI:

{
  "mcpServers": {
    "byjg-docs": {
      "httpUrl": "https://mcpdocs.byjg.com/mcp"
    }
  }
}

gemini mcp add --transport http --scope user \
  --header "Authorization: Bearer <TOKEN>" \
  byjg-docs https://your-server.example.com/mcp

Or edit ~/.gemini/settings.json directly. Note the key is httpUrl; url means the older SSE transport in Gemini CLI:

{
  "mcpServers": {
    "byjg-docs": {
      "httpUrl": "https://your-server.example.com/mcp",
      "headers": {
        "Authorization": "Bearer <TOKEN>"
      }
    }
  }
}

Verify: gemini mcp list.

Qwen Code

Qwen Code is a fork of Gemini CLI and reads the same format, from ~/.qwen/settings.json:

No token
With token

qwen mcp add --transport http --scope user --trust \
  byjg-docs https://mcpdocs.byjg.com/mcp

Or edit ~/.qwen/settings.json directly. As in Gemini CLI, the key is httpUrl; url means the older SSE transport:

{
  "mcpServers": {
    "byjg-docs": {
      "httpUrl": "https://mcpdocs.byjg.com/mcp",
      "trust": true
    }
  }
}

qwen mcp add --transport http --scope user --trust \
  --header "Authorization: Bearer <TOKEN>" \
  byjg-docs https://your-server.example.com/mcp

Or edit ~/.qwen/settings.json directly. As in Gemini CLI, the key is httpUrl; url means the older SSE transport:

{
  "mcpServers": {
    "byjg-docs": {
      "httpUrl": "https://your-server.example.com/mcp",
      "trust": true,
      "headers": {
        "Authorization": "Bearer <TOKEN>"
      }
    }
  }
}

trust skips the confirmation prompt on every tool call (in a trusted workspace). The tools only read public documentation, so there is nothing to confirm; drop it if you would rather approve each call.

Verify: qwen mcp list, or /mcp inside a session.

Kilo CLI

Kilo CLI names the key mcp and marks a remote server with "type": "remote". Add to ~/.config/kilo/kilo.json (or kilo.jsonc):

No token
With token

{
  "mcp": {
    "byjg-docs": {
      "type": "remote",
      "url": "https://mcpdocs.byjg.com/mcp",
      "enabled": true
    }
  },
  "permission": {
    "byjg-docs_*": "allow"
  }
}

{
  "mcp": {
    "byjg-docs": {
      "type": "remote",
      "url": "https://your-server.example.com/mcp",
      "enabled": true,
      "headers": {
        "Authorization": "Bearer {env:BYJG_DOCS_TOKEN}"
      }
    }
  },
  "permission": {
    "byjg-docs_*": "allow"
  }
}

{env:BYJG_DOCS_TOKEN} keeps the token out of the file: export it before starting Kilo. A literal "Bearer <TOKEN>" works too.

Kilo exposes the tools as byjg-docs_search_docs, byjg-docs_get_document and so on; the byjg-docs_* rule lets them run without asking. If your config already has "*": "ask", put the rule after it -- the last matching rule wins.

Verify: kilo mcp list.

OpenCode

Kilo CLI is a fork of OpenCode, and OpenCode reads the same mcp key with "type": "remote". Add it to the project, or with --global to ~/.config/opencode/opencode.json:

No token
With token

opencode mcp add byjg-docs --global --url https://mcpdocs.byjg.com/mcp

Or edit opencode.json (or opencode.jsonc) directly:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "byjg-docs": {
      "type": "remote",
      "url": "https://mcpdocs.byjg.com/mcp",
      "enabled": true
    }
  }
}

opencode mcp add byjg-docs --global --url https://your-server.example.com/mcp \
  --header "Authorization=Bearer <TOKEN>"

Or edit opencode.json (or opencode.jsonc) directly:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "byjg-docs": {
      "type": "remote",
      "url": "https://your-server.example.com/mcp",
      "enabled": true,
      "headers": {
        "Authorization": "Bearer {env:BYJG_DOCS_TOKEN}"
      }
    }
  }
}

{env:BYJG_DOCS_TOKEN} keeps the token out of the file: export it before starting OpenCode. A literal "Bearer <TOKEN>" works too.

OpenCode 2.x writes this as mcp.servers.byjg-docs and reads both shapes. MCP tools run without asking by default, so no permission rule is needed.

Verify with a prompt that calls the tool:

opencode run "Search the byjg-docs MCP server for MicroOrm and reply with the first URL"

opencode mcp list asks the background service, which may not have loaded the new config yet.

Cursor

Click the gear icon at the top of the chat panel, open Tools & MCP, and click Add Custom MCP. That opens ~/.cursor/mcp.json (available in every project; use <project-root>/.cursor/mcp.json for one project only):

No token
With token

{
  "mcpServers": {
    "byjg-docs": {
      "url": "https://mcpdocs.byjg.com/mcp"
    }
  }
}

{
  "mcpServers": {
    "byjg-docs": {
      "url": "https://your-server.example.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:BYJG_DOCS_TOKEN}"
      }
    }
  }
}

${env:BYJG_DOCS_TOKEN} keeps the token out of the file: export it before starting Cursor. A literal "Bearer <TOKEN>" works too.

The server shows up under Tools & MCP with a green dot once connected.

VS Code

Click the gear icon at the top of the Chat view, open MCP Servers, and click Add Server -- or run MCP: Open User Configuration from the Command Palette. Either way you edit mcp.json in your user profile.

VS Code's format differs from the others: the top-level key is servers, and each entry declares its type:

No token
With token

{
  "servers": {
    "byjg-docs": {
      "type": "http",
      "url": "https://mcpdocs.byjg.com/mcp"
    }
  }
}

{
  "servers": {
    "byjg-docs": {
      "type": "http",
      "url": "https://your-server.example.com/mcp",
      "headers": {
        "Authorization": "Bearer <TOKEN>"
      }
    }
  }
}

For a single workspace the same content goes in .vscode/mcp.json -- but that file is usually committed, so keep the token out of it and use an input variable instead.

The server appears under Installed in the MCP Servers view.

JetBrains IDEs

Applies to IntelliJ IDEA, PhpStorm, PyCharm, WebStorm and the rest, through AI Assistant. Remote HTTP servers need version 2025.3 or later.

Open Settings > Tools > AI Assistant > Model Context Protocol (MCP), click + (Add server), select the HTTP tab and paste:

No token
With token

{
  "mcpServers": {
    "byjg-docs": {
      "url": "https://mcpdocs.byjg.com/mcp"
    }
  }
}

{
  "mcpServers": {
    "byjg-docs": {
      "url": "https://your-server.example.com/mcp",
      "headers": {
        "Authorization": "Bearer <TOKEN>"
      }
    }
  }
}

If your version does not send the header (the status shows an authentication error), use the STDIO tab with the same mcp-remote bridge as Claude Desktop.

Leave Server level on Global so it is available in every project (saved to ~/.ai/mcp/mcp.json), and click OK. The Status column turns green once connected.

The tools

`search_docs(query, limit, category, project)`

The main entry point. Returns ranked passages, each with its source path, public URL and project.

query -- natural language or an exact symbol name; both work
limit -- defaults to 8, capped at 25
category -- php, devops, js, ai, helm, ...
project -- micro-orm, restserver, docker-easy-haproxy, ...

`get_document(source_path)`

Full markdown of one page, addressed by the source a search result reports (php/micro-orm/active-record.md). Use when a hit is relevant but truncated.

`list_projects()`

Inventory grouped by category and project, with counts. Lets the model discover what exists before searching.

Troubleshooting

Symptom	Cause
`401 Unauthorized`	The server you pointed at requires a token (`BYJG_DOCS_AUTH_TYPE=bearer`); `mcpdocs.byjg.com` does not. Use the With token tabs, and check the header is `Authorization: Bearer <TOKEN>`
`404 Not Found`	The URL is missing the `/mcp` path
Timeout / connection refused	The server is unreachable; check `https://mcpdocs.byjg.com/healthz` in a browser
Works in one directory only (Claude Code)	Registered with the default `local` scope; re-add with `--scope user`
`npx: command not found` (Claude Desktop, JetBrains STDIO)	Node.js is not installed, or not on the `PATH` the app sees

Claude Code​

Claude Desktop​

Codex CLI​

Gemini CLI​

Qwen Code​

Kilo CLI​

OpenCode​

Cursor​

VS Code​

JetBrains IDEs​

The tools​

search_docs(query, limit, category, project)​

get_document(source_path)​

list_projects()​

Troubleshooting​