Find and update organization contact details from their website using Anthropic Claude
Introduction
An AI agent can browse an organization's website to automate the tedious process of keeping their contact details updated.
In this example we'll use Anthropic's Claude Haiku 3.5 model. You can use any model from any provider that supports tools.
The agent is given two llm.tool: tools:
| fetch_url | Return the content of a URL in Markdown format. |
| org_update | Update an organization record with any combination of: mailing address, city, state/province, postal code, country, and phone number. |
The fetch_url tool uses Cerb's built-in http.request: automation command and |html_to_text filter to crawl websites at no extra cost.
You can optionally use a service like Tavily or Crawl4AI for a more sophisticated solution. Crawl4AI uses an actual browser engine to process content generated by Javascript.
Some websites block automated crawlers with scripting or CAPTCHAs.
Installation
Connect to an Anthropic account
Create an Anthropic connected account in Cerb.
Import the example workflow
Create a new workflow from Search » Workflows » (+) » Empty.
Copy and paste the following workflow template:
workflow:
name: wgm.ai_agent.org_update
version: 2025-05-13T00:00:00Z
description: Automatically update organization contact details from their website using Anthropic Claude.
website: https://cerb.ai/resources/workflows/
requirements:
cerb_version: >=11.0 <11.2
cerb_plugins: cerberusweb.core,
config:
chooser/anthropic_account_id:
label: Anthropic Account:
record_type: connected_account
record_query: anthropic
records:
automation/scriptUpdateOrg:
fields:
name: wgm.ai_agent.org_update.interaction
extension_id: cerb.trigger.interaction.worker
description@text:
script@raw:
inputs:
record/org:
required@bool: yes
record_type: org
expand: customfields
start:
set:
config@json: {{cerb_workflow_config('wgm.ai_agent.org_update')|json_encode}}
tool_labels:
fetch_url@raw: Fetching URL: {{url}}
org_update@raw: Updating org contact info
# Prompt for company website if we don't have one on file
outcome/noWebsite:
if@bool: {{not inputs.org.website}}
then:
data.query/hosts:
output: results
inputs:
query@text:
type:worklist.subtotals
of:address
by:[host]
query:(
org.id:{{inputs.org.id}}
)
format:dictionaries
await:
form:
title: Update Organization
elements:
text/prompt_url:
label: What is the organization's website URL?
required@bool: yes
type: url
default: https://www.{{results.data|first.host}}
placeholder: https://www.example.com
# Update the org record
record.update:
output: org_record
inputs:
record_type: org
record_id: {{inputs.org.id}}
fields:
website: {{prompt_url}}
on_success:
var.set:
inputs:
key: inputs:org:website
value: {{prompt_url}}
await/prepare:
form:
title: Update Organization
elements:
submit:
is_automatic@bool: yes
# Ask an AI agent to find the contact details on their website
llm.agent:
output: llm_results
inputs:
llm:
anthropic:
model: claude-3-5-haiku-latest
authentication: cerb:connected_account:{{config.anthropic_account_id}}
system_prompt@text:
You are a helpful AI assistant.
Your job is to fetch contact details from company websites and update the internal CRM system.
Use your tools to browse the website and to update records.
You may need to fetch multiple URLs to find the "Contact" or "About us" page.
You're a "one shot" agent, so you don't need to ask a follow-up question.
messages:
message:
role: user
content@text:
Find the company address and telephone number for:
Org: {{inputs.org.name}}
ID: {{inputs.org.id}}
URL: {{inputs.org.website}}
tools:
automation/fetch_url:
uri: cerb:automation:wgm.ai_agent.org_update.tool.fetch_url
automation/org_update:
uri: cerb:automation:wgm.ai_agent.org_update.tool.update_info
on_tool:
await:
form:
elements:
llmTranscript/prompt_transcript:
session_id: {{llm_results.session_id}}
tool_labels@key: tool_labels
submit:
is_automatic@bool: yes
await/results:
form:
title: Results
elements:
llmTranscript/prompt_transcript:
session_id: {{llm_results.session_id}}
tool_labels@key: tool_labels
policy_kata@raw:
commands:
data.query:
deny/type@bool: {{query.type != 'worklist.subtotals'}}
allow@bool: yes
llm.agent:
allow@bool: yes
record.update:
deny/type@bool: {{inputs.record_type is not record type ('org')}}
allow@bool: yes
automation/toolFetchUrl:
fields:
name: wgm.ai_agent.org_update.tool.fetch_url
extension_id: cerb.trigger.llm.tool
description: Fetch a website URL as Markdown
script@raw:
inputs:
text/url:
required@bool: yes
type: url
description: A website URL in the format of: https://example.com/
start:
log: Fetching URL: {{inputs.url}}
http.request/get:
output: http_response
inputs:
method: GET
url: {{inputs.url}}
headers:
User-Agent: Cerb
on_error:
on_success:
return:
content@text:
{{http_response.body|html_to_text(truncate=50000)}}
policy_kata@raw:
commands:
http.request:
deny/method@bool: {{inputs.method not in ['GET']}}
allow@bool: yes
automation/toolOrgUpdateInfo:
fields:
name: wgm.ai_agent.org_update.tool.update_info
extension_id: cerb.trigger.llm.tool
description: Update contact info on an organization
script@raw:
inputs:
text/org_id:
required@bool: yes
description: The organization record ID (e.g. 1).
type: number
text/street:
description: The optional street address on multiple lines without city, state, zip, country.
type: freeform
text/city:
description: The optional city.
type: freeform
text/postal:
description: The optional postal code.
type: freeform
text/country:
description: The optional country.
type: freeform
text/province:
description: The optional state or province.
type: freeform
text/phone:
description: The optional main phone number.
type: freeform
# text/region:
# description: The optional region based on the city and country.
# type: freeform
# allowed_values@list:
# North America
# South America
# Europe
# Middle East
# Asia
# Africa
# Australia & New Zealand
start:
record.update/org:
output: updated_org
inputs:
record_type: org
record_id@key: inputs:org_id
fields:
street@key,optional: inputs:street
city@key,optional: inputs:city
country@key,optional: inputs:country
phone@key,optional: inputs:phone
postal@key,optional: inputs:postal
province@key,optional: inputs:province
#region@key,optional: inputs:region
updated@date: now
on_error:
return:
content: There was an error updating the organization record.
on_success:
return:
content: Organization record updated successfully.
policy_kata@raw:
commands:
record.update:
deny/type@bool: {{inputs.record_type is not record type ('org')}}
allow@bool: yes
card_widget/cardOrgActions:
fields:
name: Actions
record_type: cerberusweb.contexts.org
extension_id: cerb.card.widget.form_interaction
pos: 100
width_units: 4
zone: content
extension_params:
interactions_kata@text:
interaction/updateContactInfo:
label: Update Contact Info
uri: cerb:automation:wgm.ai_agent.org_update.interaction
icon: address-book
inputs:
org@key: record_id
after:
refresh_widgets@list: Properties
is_popup: 1Click the Continue button. Link your Anthropic account and click Continue again.
The following records will be created:
Usage
From Search » Organizations, open an organization card popup.
Click the new Update Contact Info button at the bottom of the card.
If you haven't added a website URL to the contact you'll be prompted to enter it. If email addresses are linked to the organization, the most common hostname will be used as the default website.
The AI agent will browse the website to find contact details. It can follow a sequence of likely links (e.g. contact, about) to discover contact information.
The record is now automatically updated with the current contact info:
This also works in non-English languages.
