Skip to main content

MCP Server

Rostyman includes a built-in MCP server that allows AI tools to interact with your API workspace. When running, AI assistants such as Claude Desktop, Claude Code, Cursor, Windsurf, and VS Code Copilot can browse your collections, send requests, manage variables, and more — all through natural language.

Enabling the MCP Server

The MCP server starts automatically when Rostyman launches. To check its status or configure it:

  1. Open Settings (Ctrl+,) → IntegrationsMCP Server
  2. The status indicator shows whether the server is running, the protocol (HTTP or HTTPS), and the current URL

You can also see server status in:

  • The status bar at the bottom of the window (green dot + client count)
  • The MCP panel in the right sidebar (Overview tab)

MCP Panel — Overview Tab

The Overview tab in the MCP panel shows the live server state.

Server Card

ElementDescription
Status indicatorGreen border = running, default border = stopped
Running status"Running — localhost:3100 · 2 clients" or "Stopped"
Start / Stop buttonStarts or stops the server
Settings buttonOpens Settings → Integrations → MCP Server
Register with Claude DesktopAuto-registers Rostyman in Claude Desktop's config file
Copy ConfigCopies the MCP JSON configuration to the clipboard

Connected Clients Tab

Lists AI agents currently connected via SSE:

FieldDescription
NameFriendly name parsed from user-agent (Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, MCP Remote Bridge, Node.js Client, Python Client)
IPSource IP address of the connection
SinceDuration since connection was established (e.g. "2m ago")
DisconnectForce-closes the client's SSE stream

Click Refresh to reload the client list. The list updates automatically when clients connect or disconnect.

Connected Servers Tab

Lists external MCP servers Rostyman is connected to as a client, with status and tool count. See MCP Client for details.


MCP Panel — Servers Tab

Manages external MCP servers (client connections). See MCP Client.


MCP Panel — Logs Tab

Controls

ControlAction
Search inputFilter log lines in real time
Verbose checkboxToggle full request/response payloads in logs
Clear LogsClears the in-memory ring buffer (500 entries max)
Open folder iconOpens the log directory in the file manager

Log Entry Format

Each line shows: time LEVEL [source] message

  • ERROR — red
  • WARN — amber
  • TOOL — accent color (tool call events with timing)
  • INFO — muted

Log Content

Every log entry captures:

  • HTTP requests: method, URL, IP address, session ID
  • JSON-RPC method calls: initialize, tools/list, tools/call, etc.
  • Tool calls: tool name, arguments, result preview, elapsed time (e.g. "ok (42ms)")
  • Session lifecycle: initialized, closed, client disconnected
  • Auth failures and rate limit hits

Logs are saved to date-based files in {userData}/logs/mcp/. Files older than 30 days are automatically deleted. The stats bar at the bottom shows the file count and total size.


Settings Reference

All settings are in Settings (Ctrl+,) → IntegrationsMCP Server.

Port

Default: 3100. To change: edit the port field and press Enter or click away. The server restarts on the new port. If the port is busy, Rostyman automatically tries the next available port.

Access Mode

ModeDescription
Local only (default)Accepts connections from localhost only. CORS restricted to localhost origins. No auth required.
NetworkAccepts connections from any device on your network. Requires a Bearer token. Rate limited to 60 req/min per IP.

Auth Token (Network Mode)

Rostyman generates a cryptographically secure token when Network mode is enabled.

ActionHow
View tokenShown partially masked in Settings
Copy tokenClick the copy icon
RegenerateClick the refresh icon — creates a new token and invalidates the old one

AI tools must include the token in every request:

Authorization: Bearer your-token-here

TLS / HTTPS

Enable HTTPS for secure connections. Required for Claude.ai web connectors and recommended for Network mode.

Prerequisites: Install mkcert

# Windows
choco install mkcert
# or
scoop install mkcert

# macOS
brew install mkcert

# Linux
# See https://github.com/FiloSottile/mkcert#installation

Setup:

  1. Rostyman auto-detects mkcert. Click Recheck if it was just installed.
  2. Click Generate with mkcert to create locally-trusted certificates.
  3. Toggle TLS on — the server restarts with HTTPS.

Certificates are stored in {userData}/certs/. The server URL and all config snippets automatically switch to https://.

Tunnel Access

Expose your local server to the internet so remote AI tools (including Claude.ai web) can connect.

ngrok:

  1. Install ngrok from ngrok.com
  2. Rostyman detects it automatically — a Copy ngrok command button appears
  3. Run the copied command in a terminal
  4. Use the generated https://xxx.ngrok-free.app/mcp URL in your AI tool's config

Cloudflare Tunnel:

cloudflared tunnel --url http://localhost:3100

Use the generated URL in your AI tool's config.

When using a tunnel, keep TLS off — the tunnel provider handles HTTPS at their end.


Connecting AI Tools

  1. In Rostyman, click Register with Claude Desktop (in the MCP panel Overview tab or in Settings)
  2. Restart Claude Desktop

Claude Desktop (Manual Config)

Paste into Claude Desktop's config file:

  • Windows: %APPDATA%\Claude\claude_desktop_config.json
  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Linux: ~/.config/claude/claude_desktop_config.json
{
"mcpServers": {
"rostyman": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://localhost:3100/mcp"]
}
}
}

Restart Claude Desktop after editing the config.

Claude.ai Web

  1. Set up a tunnel pointing to port 3100 (ngrok or cloudflared)
  2. Go to claude.ai → Settings → Connectors → +
  3. Enter the tunnel URL (e.g. https://xxx.ngrok-free.app/mcp)
  4. All 18 tools appear in the connector's tool list

Copy Config (All Other Tools)

Click Copy Config in the MCP panel to get the JSON snippet formatted for the currently running server (HTTP or HTTPS, correct port).


Available Tools

Rostyman exposes 18 tools in three categories, visible in the Show Tools dialog (accessible from the MCP panel Overview tab or Settings).

Read Tools (8)

ToolDescriptionParameters
list_collectionsLists all collections in the active workspace
get_collectionGets collection details with all folders and requestscollectionId
get_requestGets full request details (URL, headers, body, auth)requestId
search_requestsSearches requests by name, URL, or method (up to 50 results)query
list_environmentsLists all environments for a collectioncollectionId
get_environmentGets environment variables (secret values are masked as ••••••)environmentId
get_globalsGets all global variables for the workspace
get_historyGets the last 20 executions of a request with status and timingrequestId

Execute Tools (2)

ToolDescriptionParameters
send_requestSends a saved request (by ID) or an ad-hoc request (by method + URL). Resolves variables before sending.requestId?, method?, url?, headers?, body?
run_collectionRuns all requests in a collection or specific folder, resolves variables, returns a summarycollectionId, folderId?, environmentId?

Write Tools (3)

ToolDescriptionParameters
create_collectionCreates a new empty collectionname
create_requestCreates a request in a collectioncollectionId, name, method, url, folderId?, headers?, bodyMode?, bodyContent?
set_variableSets a variable in environment, collection, or global scope (upserts)scope (environment/collection/global), scopeId, key, value

Tools Dialog

Click Show Tools in the Overview tab or Settings to open a searchable dialog showing all tools. Each tool entry displays:

  • Tool name (monospace)
  • Category badge (Read / Execute / Write)
  • Description
  • Parameter names (required params in secondary color, optional params ending in ? in muted color)

Resources (4)

In addition to tools, Rostyman exposes 4 MCP resources that AI tools can read directly:

Resource URIDescription
rostyman://collectionsJSON list of all collections (id + name) in the active workspace
rostyman://environmentsAll environments across all collections
rostyman://globalsAll global variables for the workspace
rostyman://historyRecent request history entries

Resources are read-only and reflect the current workspace state.


Variable Resolution

When send_request or run_collection is called, variables are resolved in this order (highest priority last):

  1. Global variables
  2. Collection variables
  3. Environment variables (from the collection's active environment, or the environmentId parameter)

Built-in variables like {{$timestamp}} and {{$randomInt}} are also resolved.


Troubleshooting

Port already in use

Rostyman automatically tries the next available port. You can also manually change the port in Settings.

Claude Desktop shows no tools

  1. Verify Rostyman is running and the MCP server is started (green dot in status bar)
  2. Confirm the port matches in both apps
  3. Restart Claude Desktop after any config change
  4. Check the Logs tab for connection or initialization errors

Auth errors in Network mode

  1. Confirm the token in Rostyman matches the one in your AI tool's config
  2. The header format must be exactly: Authorization: Bearer your-token-here
  3. Regenerate the token in Settings if needed and update the AI tool config

Connection drops after idle

In Network or tunnel mode, SSE connections may time out if idle for too long. The AI tool will reconnect automatically on the next request. This is normal behavior.