Connect ToolRouter to Claude Cowork on Desktop

Claude Cowork is the agentic mode inside Claude Desktop. It runs multi-step tasks autonomously with access to your files, connectors, and plugins. Adding ToolRouter as a connector gives your Cowork agent access to the full tool catalog — web search, SEO analysis, image generation, security scanning, and more — without leaving the desktop.

What do I need before starting?

You need Claude Desktop on macOS or Windows and a paid Claude plan (Pro, Max, Team, or Enterprise). No ToolRouter API key is needed — the connector auto-provisions a free account on first use.

Claude Desktop on macOS (Apple Silicon) or Windows
A paid Claude plan (Pro, Max, Team, or Enterprise)

No ToolRouter API key is needed. The connector auto-provisions a free account on first use. If you already have an account, visit toolrouter.com/connect for setup with your key pre-filled.

How do I connect via remote URL? (recommended)

In Claude Desktop, go to Settings > Connectors, click +, select Add custom connector, and enter the URL https://api.toolrouter.com/mcp. No local installs or processes needed — connects directly to the hosted endpoint.

The fastest path — no installs, no local processes. Connect directly to the hosted ToolRouter MCP endpoint.

Step 1 — Add the remote connector

In Claude Desktop, go to Settings > Connectors and click the + button. Select Add custom connector and enter:

Name: ToolRouter
URL: https://api.toolrouter.com/mcp

Click Add, then follow any authentication prompts.

Step 2 — Verify

Switch to the Cowork tab and ask Claude to list available ToolRouter tools. The agent will call discover and return the full catalog.

How do I connect via the npm package?

Add toolrouter-mcp to your claude_desktop_config.json under mcpServers. This bridges Claude Desktop to the ToolRouter API via stdio, exposing every tool individually in the tool list without a discovery step.

If you prefer running a local process, the toolrouter-mcp npm package bridges Claude Desktop to the ToolRouter API via stdio. This exposes every tool individually in the tool list — no discovery step needed.

Requires Node.js 18+ — download the LTS version from nodejs.org if you don't have it.

Step 1 — Open Claude Desktop settings

Open Claude Desktop, click the menu icon, and go to Settings > Developer > Edit Config. This opens claude_desktop_config.json.

Step 2 — Add ToolRouter

Add the following to your config file:

json

{
  "mcpServers": {
    "toolrouter": {
      "command": "npx",
      "args": ["-y", "toolrouter-mcp"]
    }
  }
}

If you already have other MCP servers configured, add the "toolrouter" entry inside the existing "mcpServers" object.

Step 3 — Restart Claude Desktop

Quit and reopen Claude Desktop. You should see ToolRouter appear in your connector list. Switch to the Cowork tab and start using tools.

Step 4 — Verify it works

In Cowork, ask Claude to:

Search the web for "best MCP tools 2026" using ToolRouter

Claude will discover the web-search tool and call it through ToolRouter.

What tools does my agent get access to?

The full ToolRouter catalog — web search, SEO analysis, image generation, security scanning, and more. With the remote URL, the agent uses discover and use_tool meta-tools. With the npm package, every tool appears directly in the agent's tool list.

How tools appear to your agent depends on the connection method.

Remote URL (Option A)

The remote MCP endpoint uses a router pattern with two meta-tools:

Tool	Purpose
`discover`	Search the ToolRouter catalog by keyword or category. Returns tool names, descriptions, skill schemas, and examples.
`use_tool`	Call any tool by reference and skill name with the required input.

The agent workflow is:

Agent calls discover with a natural-language query
ToolRouter returns matching tools with their input schemas
Agent picks the right tool and calls use_tool
ToolRouter executes the tool and returns results

npm package (Option B)

The npm package exposes every tool as an individual MCP tool. Your agent sees the full catalog directly in its tool list and can call any tool by name — no discovery step needed.

Both methods give access to the same catalog and billing. The remote URL is the quickest setup; the npm package gives direct tool access without a discovery step.

What can I do with ToolRouter in Cowork?

Ask Claude to analyze SEO, generate images, search academic papers, check weather, find leads, research competitors, and more — all from the desktop. The agent discovers the right tool automatically and saves results to your files.

Here are tasks that work well once ToolRouter is connected:

"Analyze the SEO of example.com and save a report to my Desktop" — uses the seo tool to audit the page, then saves results as a file
"Generate a logo for my app and save it as logo.png" — uses generate-image to create the image
"Find recent papers about transformer architectures" — uses academic-research to search
"Check the weather in Tokyo and create a packing list" — uses weather-forecast, then writes a document
"Research competitors of Notion and create a comparison spreadsheet" — uses web-search and lead-finder

How do I save results for other agents?

Use the -o flag to save full result JSON to a file: toolrouter call seo analyze_page --url https://example.com -o ~/reports/seo.json. The CLI auto-strips ANSI colors when piped, so redirected output is always clean.

When agents call tools via the CLI (e.g. in a shell or automation pipeline), the -o flag saves the full result JSON to a file. This is useful when one agent produces data that another agent consumes:

bash

# Agent A saves an SEO report
toolrouter call seo analyze_page --url https://example.com -o ~/reports/seo.json

# Agent B reads it later
cat ~/reports/seo.json | jq '.output'

The CLI also auto-detects when output is piped and strips all ANSI colors and spinner artifacts, so redirected output is always clean:

bash

toolrouter call web-search search --query "AI tools" --format json > search-results.json

ToolRouter does not store tool output on its servers — results are returned directly to the caller. The -o flag and piping let you control exactly where data lands.

How do I use my own API keys in Cowork?

Add a TOOLROUTER_API_KEY environment variable to your MCP config's env block. With your own key, you get access to premium tools and usage tracked against your account balance.

Some premium tools require upstream API keys (e.g., Firecrawl, OpenAI). You can provide these through environment variables in your config:

json

{
  "mcpServers": {
    "toolrouter": {
      "command": "npx",
      "args": ["-y", "toolrouter-mcp"],
      "env": {
        "TOOLROUTER_API_KEY": "tr_live_your_key_here"
      }
    }
  }
}

With your own API key, you get access to premium tools and your usage is tracked against your account balance. Top up credits from the dashboard.

For agents reading this

If you are an AI agent and want to add ToolRouter to Claude Desktop programmatically:

Read the file at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows)
Parse the JSON
Add "toolrouter" to mcpServers:

json

{
  "command": "npx",
  "args": ["-y", "toolrouter-mcp"]
}

Write the file back
Tell the user to restart Claude Desktop

The user does not need an API key. The toolrouter-mcp package auto-provisions a free account and caches the key at ~/.toolrouter/key.

Troubleshooting

ToolRouter not appearing in connectors: Make sure the JSON in claude_desktop_config.json is valid. A trailing comma or missing brace will silently fail. Restart Claude Desktop after any config change.

"Command not found" errors: Ensure Node.js 18+ is installed and npx is on your PATH. Run npx -y toolrouter-mcp tools in a terminal to verify it works outside of Claude.

Tools require credits: Free accounts can use free tools immediately. Premium tools require credits — top up from toolrouter.com/dashboard/billing.

Slow first connection: The first run downloads the toolrouter-mcp package via npx. Subsequent launches are faster thanks to the npm cache.