Workflow Automation
AdsCrawl for Paperclip
Give the agents you manage in Paperclip access to live web pages through AdsCrawl. Connect the hosted MCP server, assign a research task and keep the findings with the task. Paperclip coordinates the work; AdsCrawl runs the browser session.
What you can do
Web research in tasks
Ask an assigned agent to read a public page and report its heading, findings and source URL in the Paperclip task.
Browser actions across steps
Use the MCP tools to open, read and navigate pages in one hosted session, then close it when the task finishes.
Repeatable research workflows
After a manual task succeeds, reuse its instructions in a Paperclip routine for recurring page checks and reports.
Before you start
- An existing Paperclip instance, a configured agent and permission to manage its connections. The connector steps follow Paperclip’s v2026.1106.0 documentation; labels depend on your installed version.
- An AdsCrawl API key with available credits, plus outbound HTTPS access to api.adscrawl.net from the environment making MCP calls.
- For the optional runtime method: an existing claude_local agent using the Claude CLI engine, with Claude Code installed and authenticated on its execution host.
SETUP METHOD 01
MCP connector
Recommended when you want to choose agent access and action permissions in Paperclip. Add AdsCrawl as a custom remote MCP server.
Add the server URL
Open Connectors, choose Connect your own MCP server and enter the endpoint below. Name the connection AdsCrawl. It uses Streamable HTTP.
https://api.adscrawl.net/mcpSet the API key header
Use the API-key/custom-header setup and enter your actual key in the credential form as x-api-key. The value below is a placeholder. Keep the key in the connection credentials rather than task text.
x-api-key: YOUR_ADSCRAWL_API_KEYChoose agents and tools
Choose which agents can use this connection, then open Permissions and refresh the action list. For the first page-reading task, make browser_open, browser_read, browser_navigate and browser_close available. Configure any other actions to match the intended workflow.
Run a first task
Create a Paperclip task, assign it to an eligible agent and use the prompt below. Expect the heading Example Domain, a source URL and browser_close with confirmed: true. Inspect the task’s run output if a tool is missing or a permission blocks it.
SETUP METHOD 02
Claude Code runtime MCP
An alternative for an existing claude_local agent using engine: cli. The runtime connects directly; these calls do not pass through Paperclip’s connector action permissions or review queue. Choose one method for the same agent.
Locate the execution workspace
Check the agent’s actual adapter, engine and cwd. This recipe targets the legacy claude_local adapter with the CLI engine, not the newer Paperclip Runner. Work in that cwd on the host or sandbox where the agent runs.
Merge the project MCP configuration
Add this adscrawl entry to .mcp.json in that workspace, preserving existing servers. The environment-variable reference is expanded by Claude Code; no actual key belongs in this file.
{ "mcpServers": { "adscrawl": { "type": "http", "url": "https://api.adscrawl.net/mcp", "headers": { "x-api-key": "${ADSCRAWL_API_KEY}" } } } }Supply the execution environment
Set ADSCRAWL_API_KEY in the agent’s adapter env using a Paperclip secret reference. A variable set only in your own terminal may not reach a service or sandbox. Other agents sharing this project configuration can also discover the server.
Verify from the same environment
Run claude mcp list from the same cwd and environment and confirm AdsCrawl connects. Complete any project MCP trust setup required by Claude Code before a headless run. Then assign the first task below and confirm that its browser session closes.
claude mcp list
Try it
After setup, send this to your agent to check that the integration works.
Use AdsCrawl to open https://example.com and read the page heading. If navigation is needed, navigate within the same session. Record the heading and source URL in this Paperclip task. Close the browser with browser_close even if reading fails, and report whether confirmed is true. Do not take a screenshot for this check.Usage & limits
- This is a custom MCP integration guide. Paperclip manages tasks and agent execution; you configure the agent and its model separately.
- AdsCrawl sessions consume your AdsCrawl credits. Check AdsCrawl Billing & Usage for browser charges; do not assume Paperclip’s reported model costs include them.
- Keep browser_close available and check confirmed: true. Ending a Paperclip task or a runtime process does not confirm that the remote browser session has stopped.
- browser_screenshot creates a publicly accessible JPEG link. Enable it for workflows where the image can be shared; it is not needed for the first reading task.