Launched:self-serve migrations intoSuperhuman Docs (Coda)
Try it now
01Agent-first
Runs where you already work
Plug it into Claude, ChatGPT or Cursor. Describe the move in plain English; the agent runs it.
02Engineer-led
Our production engine, unlocked
The pipeline our engineers use on managed enterprise migrations — the same code, now something you can drive yourself.
03Pricing
Try 10 pages free, then $1 a page
Credit-based, pay-as-you-go. No scoping call, no quote — sample it on your own docs before you spend anything.
04Sources
NotionSlabConfluenceSoonGoogle DocsSoon
Skip to content

Can an AI Agent Migrate Your Help Desk? (Zendesk, Intercom, Freshdesk)

AI agents can draft mappings and review samples, but a help desk migration needs import endpoints, archived-ticket exports and validation that vendor MCP servers were not built to provide. Here is what works.

Roopendra Talekar Roopendra Talekar · · 13 min read
Can an AI Agent Migrate Your Help Desk? (Zendesk, Intercom, Freshdesk)
TALK TO AN ENGINEER

Planning a migration?

Get a free 30-min call with our engineers. We'll review your setup and map out a custom migration plan — no obligation.

Schedule a free call
  • 1,500+ migrations completed
  • Zero downtime guaranteed
  • Transparent, fixed pricing
  • Project success responsibility
  • Post-migration support included

Can an AI Agent Migrate Your Help Desk? (Zendesk, Intercom, Freshdesk)

An AI agent, whether Claude, ChatGPT or a custom agent connected through MCP servers, can draft field mappings, review a sample migration and catch inconsistencies in your data. Running the full data move is a different job.

The model is not the obstacle. A help desk migration is a bulk, one-time, irreversible load. Vendor MCP servers are built for conversational work on a few records at a time. The paths that preserve history sit elsewhere: an admin-only import endpoint, an export that includes archived tickets, and controls that stop triggers firing on old tickets.

This guide walks through what each platform documents, so you can decide where an agent adds value and where you need a migration engine.

Last verified: 17 November 2026. Vendor MCP servers change often, so check the linked documentation before you plan a load.

Warning

Quick answer: Use the agent as an analyst, not as the migration engine. MCP servers are built for conversational work on live records. Help desk migrations need import-specific endpoints, deterministic retries and audit-grade validation.

Cross-platform constraint summary

This table maps what each vendor's official agent interface offers next to the path that preserves history.

Zendesk Intercom Freshdesk
Official agent interface Zendesk Bridge, early access from 10 November 2026, by enrollment Intercom MCP server Freshworks MCP integration, generally available from 10 September 2026
What it can write No destructive skills, bulk exports or third-party integrations during early access. Admin Center tools are read-only Internal notes, and create or update articles. No tool creates conversations or contacts 37 tools. By name, 17 create or modify data. One bulk tool updates tickets
Volume limits Layered tenant and agent limits, with numbers confirmed at enrollment Search returns up to 150 items per page 25, 50 or 100 actions per minute. 1,200, 6,000 or 12,000 actions per year
Path that sets original dates Ticket Import API, admins only REST Create Conversation accepts created_at Not covered in this guide. Check Freshdesk's API documentation
Authentication notes API tokens deactivated 30 April 2027. OAuth replaces them — MCP integration uses an API key only

Sources: Zendesk Bridge FAQ, Zendesk Ticket Import API, Zendesk API token notice, Intercom MCP, Intercom Create Conversation, Freshdesk MCP.

The pattern is the same on all three. The official agent interface is designed for conversational work, and the path that preserves history is a separate API that a migration has to call on purpose.

What goes wrong when an agent creates tickets through the standard API

A general-purpose agent reaches for the create tool in its MCP catalog, or the create endpoint it finds in public documentation. That path is designed for new work, not historical imports.

  • Timestamps reset to now. A ticket created through Zendesk's normal create endpoint is stamped with the current time, because created_at is read-only there (developer.zendesk.com). Ageing views, SLA reports and first-contact metrics break.
  • Business rules run. Zendesk says tickets created with Create Many Tickets "may be affected by your business rules, which can include sending email notifications to your end users" (developer.zendesk.com). Deleting the ticket afterwards does not recall the email. Our guide to importing tickets without sending notifications covers the controls.
  • Authorship collapses. Records are attributed to the connected user instead of the original requester or agent.
  • Archived records are missed. "Archived tickets are not included" in Zendesk's List Tickets responses (developer.zendesk.com). The run finishes, reports success, and the gap goes unnoticed without a count against the source.

The MCP control model was also designed for a different job. The MCP specification says there should always be a human in the loop who can deny tool invocations. Anthropic notes that "the autonomous nature of agents means higher costs, and the potential for compounding errors", and recommends "extensive testing in sandboxed environments" (anthropic.com). That is sound advice for agents in general. A migration makes hundreds of thousands of writes, which nobody can approve one by one.

Volume is a separate problem. Claude Code warns when MCP tool output exceeds 10,000 tokens, limits it to 25,000 tokens by default, and saves text results over 50,000 characters to a file (code.claude.com). Anthropic's engineering team recommends that agents process large data in code outside the model, because passing records through tool calls costs tokens and "models may be more likely to make mistakes when copying data between tool calls". In its example, "The agent sees five rows instead of 10,000" (anthropic.com). That is the migration-engine pattern: code moves the records, and the agent works on the plan.

Zendesk: the Ticket Import API and the traps around it

Zendesk has a dedicated Ticket Import API. It can set solved_at, updated_at and created_at on tickets, and created_at on comments. It is allowed for admins only. Zendesk's documentation says: "If a ticket with a status other than closed is imported, triggers won't run on the imported ticket. However, if the ticket is updated after import, triggers will resume." No notifications go to users CC'd on imported tickets (developer.zendesk.com).

A general-purpose agent does not reach this endpoint by default. Its create tools call the normal tickets API, which stamps the current time and runs business rules.

The specifics a migration script has to handle:

  • Users first. Tickets reference requesters, and Zendesk's import examples show each comment carrying an author_id. Users have to exist before tickets, and the loader needs a source-to-destination ID map that outlives any one session.
  • Attachment upload tokens expire in 60 minutes. Zendesk's upload endpoint returns a token that must be attached to a ticket within that window (developer.zendesk.com). Upload and attach in the same pass.
  • Archived tickets need the Incremental Ticket Export. List Tickets excludes them, and Zendesk points to the Incremental Ticket Export to get every ticket (developer.zendesk.com). Incremental exports are limited to 10 requests per minute, or 30 with the High Volume add-on (developer.zendesk.com).
  • Metrics and SLAs are not supported for imported tickets. Zendesk recommends tagging imported tickets and excluding them from reports (developer.zendesk.com).
  • The bulk import endpoint accepts up to 100 tickets per request. For 50,000 tickets, that is at least 500 requests, each needing correct requester IDs, comment authors, attachment tokens and timestamps.
  • Idempotency keys expire after two hours. A duplicate key sent two hours or more after the original request creates a new ticket (developer.zendesk.com). A purpose-built loader tracks its own state. An agent session that resumes later and relies on the old key can create duplicates.
  • Pre-1970 timestamps are rounded up. A ticket created_at or updated_at before 1970 is rounded up to 1970, and a comment created_at cannot be before 1970 or in the future (developer.zendesk.com).

For the field-by-field detail across platforms, see how to keep the created date when migrating.

Does Zendesk Bridge help with bulk migration?

Not today. Zendesk Bridge is "the trusted interaction layer between Zendesk and external AI agents". Its early access program starts 10 November 2026, by enrollment. During early access, destructive skills, bulk exports and third-party integrations are not available, and Admin Center tools are read-only. Rate limits are layered at tenant and agent level, with numbers confirmed at enrollment (support.zendesk.com).

Bridge is built for approved agents doing authorized service work under the account's own policies. That is a different job from moving years of history in one pass.

What does Zendesk's API token sunset mean for migration scripts?

Zendesk is removing API tokens as an authentication method (support.zendesk.com):

  • Accounts created on or after 28 July 2026 cannot create or use API tokens.
  • Since 27 October 2026, no account can create new API tokens.
  • On 30 April 2027, all remaining API tokens are deactivated permanently.
  • Since 28 July 2026, a token that goes unused for 30 days deactivates automatically.

Zendesk says OAuth replaces API tokens. Its notice describes creating an OAuth client in Admin Center and getting access tokens with the client_credentials or authorization_code grant. Any migration script, agent setup or integration that still authenticates with an API token stops working by 30 April 2027.

How does this change a Zendesk-to-Gorgias migration?

The destination has its own rules, and an agent does not change them. Gorgias's own Zendesk migration documentation says its native import brings in two years of history, splits Zendesk tickets with more than 250 messages into multiple tickets, and imports On-hold, Solved and Closed Zendesk tickets as Closed, because Gorgias has only Open and Closed. Whatever runs the move, someone has to decide how much history and how much status detail to carry before the first ticket moves.

For a Zendesk-specific deep dive, see Zendesk AI Agents Migration 2026: What Breaks, What Changes.

Can the Intercom MCP server create conversations or contacts?

No. Intercom's official MCP server can add an internal note to a conversation and create or update Help Center articles. The note tool "cannot send customer-visible replies", and there is no tool to create conversations or contacts.

Articles created through the MCP server are attributed to the admin who authorized the connection, and the author cannot be set per call. In a knowledge-base migration, every article appears to be written by one person.

Intercom's REST API is broader than its MCP server. The Create Conversation endpoint accepts created_at as a UTC Unix timestamp. Intercom recommends the field only for migrating past conversations from another source into Intercom, and uses the current time if it is not provided (developers.intercom.com). So Intercom can hold historical conversations. A general-purpose agent using the MCP server cannot create them.

A script that uses the REST API directly still has ordering work to do. Each conversation's from object identifies a contact by id, user_id or email (developers.intercom.com). Contacts have to be loaded and mapped first, then conversations.

The MCP server supports US and EU hosted Intercom workspaces. AU hosted workspaces are not yet supported (developers.intercom.com).

How far does Freshdesk's MCP allowance stretch for a migration?

Not far. Freshdesk's MCP integration has been generally available since 10 September 2026. It lists 37 tools. By name, 20 read data and 17 create or modify it. Its one bulk tool, createTicketBulkUpdate, updates multiple tickets. Authentication is by API key only.

The limits:

Plan Actions per minute Included actions per year
Growth 25 1,200
Pro 50 6,000
Enterprise 100 12,000

Source: support.freshdesk.com.

The arithmetic (ours, not Freshdesk's): at 100 actions a minute, writing 100,000 records at one record per action takes at least 1,000 minutes, before retries. The Enterprise annual allowance is 12,000 actions. Every contact, note or reply you write alongside a ticket is another action.

Freshdesk's MCP integration suits small, supervised admin tasks. It was not built as a bulk-history lane.

Security, compliance and untrusted data

A data migration needs read access to everything in the source and write access to every object in the destination. It moves customer names, emails and years of free text.

Simon Willison's "lethal trifecta" is access to private data, exposure to untrusted content, and a way to communicate externally (simonwillison.net). A migration agent has all three by design. It reads every ticket, much of that text was written by outsiders, and it holds write access to a second system. A ticket body can contain text written to look like instructions.

This is demonstrated in research, not in customer breaches. Invariant Labs showed that a malicious GitHub issue could lead an agent connected to the GitHub MCP server to copy private repository data into a public pull request (invariantlabs.ai). General Analysis showed that a support ticket containing instructions could lead an AI coding assistant connected to the Supabase MCP server, with a privileged key, to expose a secrets table (generalanalysis.com). OWASP recommends human-in-the-loop controls for privileged operations (genai.owasp.org).

Compliance adds its own constraints. Under HIPAA, a covered entity may let a business associate handle protected health information only with satisfactory assurance, in practice a business associate agreement (BAA) (law.cornell.edu). Anthropic says data sent to third parties through MCP and connectors "isn't covered under Anthropic's BAA" (privacy.claude.com). OpenAI states it does not offer a BAA for ChatGPT Business (help.openai.com).

Accountability sits with you. Anthropic's commercial terms provide the services "AS IS" (anthropic.com). OpenAI's services agreement says "Customer is solely responsible for all use of the Outputs" (openai.com). These terms are reasonable for a general-purpose tool. They also mean that if an agent-run migration corrupts your data, the clean-up is yours: finding bad records, removing them without touching live work, re-importing and repairing links.

Where AI agents genuinely help in a help desk move

AI earns its place in the planning and review phases:

  • Drafting field and status mappings. Give the agent both schemas and it produces a first-draft mapping table quickly. A person still signs off on each pair, because status and priority names rarely match one to one between platforms.
  • Reviewing a sample migration. After a sample run, an agent can compare source and destination records field by field, flag mismatches and spot patterns in the errors.
  • Writing and testing transformation code. An agent can draft the code that normalizes timestamps, converts HTML or maps statuses, and that code runs outside the model. The dlt team used this approach to migrate HubSpot to Attio. AI drafted the model, transforms and tests, humans reviewed every change, and a human caught an AI-proposed design that conflicted with GDPR deletion rules.
  • Knowledge-base drafts and duplicate suggestions. For article migrations, an agent can suggest merges, flag duplicates and draft reformatted content for human review.

The common shape of every successful AI-assisted migration: AI drafts, a person reviews, deterministic code runs the job.

Task AI agent appropriate? Why
Schema discovery and documentation Yes Read-only, low risk
First-draft field and status mappings Yes, with human review Naming differences need judgment
Generating dry-run test cases Yes Outputs are reviewed before execution
Sample review and anomaly spotting Yes Pattern matching on structured output
Full history replay via generic create endpoints No Timestamps, triggers and authorship all break
Approving its own production writes No No human in the loop for bulk decisions
Sole reconciliation report No No deterministic count verification

For preparing your data for AI agents after the move, see the CRM & Helpdesk Data Prep for AI Agents: 2026 Playbook.

Two routes for the actual data move

Once the mapping is agreed and the sample is reviewed, the data has to move through import-specific endpoints, in the right order, within rate limits and token-expiry windows. ClonePartner offers two ways to do that. You choose.

ClonePartner Express Managed migration
Who runs it You, in the browser A dedicated migration engineer who owns the project
Supported pairs Gorgias from 17 help desks. Front from Intercom or Zendesk Any source and destination, including custom fields, relational records, permissions and compliance requirements
Before you commit A free sample of 100 tickets on your real data Unlimited sample migrations, included in the price
Price Shown on each Express site One fixed price per project, starting at $500. No hourly billing, no hidden fees, no change orders
Go-live You run the full migration when the sample looks right The bulk load runs in the days before go-live while your team keeps working. Before the working day starts on Monday, the old system is frozen and a final delta migration of only changed records takes about 30 minutes

Express is a self-serve migration tool: connect your accounts, run the free sample on real data, review it, then run the full migration. The live help desk destinations are Gorgias and Front. For the technical detail of one pair, see our Zendesk to Gorgias migration guide.

A managed migration freezes the field mapping after you approve the sample and reuses it for the full run and the delta. Validation covers record-count parity, field-level checks, referential integrity and a reconciliation report. Most migrations complete in 1 to 5 business days, depending on data volume, complexity and API rate limits (larger histories take longer).

Info

If your data includes PHI, ClonePartner signs a BAA before any data moves. Controls include SOC 2 Type II and ISO 27001, GDPR, CCPA and HIPAA safeguards, AES-256 at rest, TLS 1.2+ in transit, and staging databases purged no later than 30 days after project close. Details at trust.clonepartner.com.

Six questions to ask before connecting an agent to production

If your migration plan starts with "we'll point Claude at the old help desk and let it figure it out", answer these first:

  1. Which endpoint preserves original dates and authorship? Zendesk: the Ticket Import API. Intercom: Create Conversation with created_at. Other platforms: find the path in the vendor's own documentation before you build.
  2. Which export path includes archived history? Zendesk: the Incremental Ticket Export, because List Tickets excludes archived tickets.
  3. What happens to triggers, notifications and SLAs on import? Zendesk: triggers don't run on imported tickets whose status is anything other than closed, they resume on any later update, and metrics and SLAs are not supported for imported tickets.
  4. How are attachments handled before tokens or signed URLs expire? Zendesk upload tokens must be attached within 60 minutes.
  5. What is the retry and idempotency model? Zendesk idempotency keys expire after two hours. Your loader needs its own record of what has already been written.
  6. Where is the reconciliation report? Record-count parity, field-level checks and referential integrity have to be verified outside the agent. A session log is not a reconciliation report.

If those answers are fuzzy, the agent should not be driving the load. Let AI help with the thinking, and let a migration engine or an engineer-run pipeline do the moving.

For Zendesk-specific preparation, start with our Zendesk Migration Checklist.

Info

Two ways to run it. If your source and destination are a pair ClonePartner Express supports, you can run the migration yourself: Gorgias or Front, each with a free sample on your real data first. For any other pair, or custom fields, relational records, permissions or compliance requirements, talk to a migration engineer about a fixed-price managed migration.

Frequently Asked Questions

Can Claude or ChatGPT migrate my Zendesk tickets?
Not through conversational tools or the standard create endpoint. The endpoint that sets original timestamps is Zendesk's Ticket Import API, which is admin-only and does not run triggers on imported tickets whose status is anything other than closed. Zendesk Bridge, Zendesk's layer for external AI agents, excludes bulk exports and destructive skills during early access. An agent can draft field mappings and review a sample, but the load needs purpose-built migration tooling.
What can the Intercom MCP server actually write?
Intercom's official MCP server can add internal notes to conversations and create or update Help Center articles. The note tool cannot send customer-visible replies, and no tool creates conversations or contacts. Articles are attributed to the admin who authorized the MCP connection.
How many MCP actions does Freshdesk include per year?
Freshdesk includes 1,200 MCP actions per year on Growth, 6,000 on Pro and 12,000 on Enterprise, with rate limits of 25, 50 and 100 actions per minute. At 100 actions a minute, writing 100,000 records one per action takes at least 1,000 minutes, and the Enterprise annual allowance covers 12,000 actions.
When do Zendesk API tokens stop working?
Accounts created on or after 28 July 2026 cannot create or use API tokens. No account can create new API tokens since 27 October 2026. All remaining tokens are deactivated permanently on 30 April 2027, and Zendesk says OAuth replaces them. Any script or agent setup that authenticates with an API token must move to OAuth before then.
What is the safest way to use AI in a help desk migration?
Use an AI agent to draft field and status mappings, write transformation code and tests, review sample migration results and suggest duplicates. The common pattern in successful AI-assisted migrations is that AI drafts, a person reviews, and deterministic code runs the data transfer.

More from our Blog

AI Agent Data Migration via MCP: The 2026 Guide
CRM/Help Desk/Migration Guide

AI Agent Data Migration via MCP: The 2026 Guide

AI agents are useful for drafting migration mappings and code. Vendor MCP servers are built for a few records at a time, so the bulk load needs a migration engine and someone accountable for the result.

Nachi Raman Nachi Raman · · 12 min read
Zendesk Migration Checklist
Checklist/Zendesk

Zendesk Migration Checklist

A technical checklist for migrating to Zendesk — covering API rate limits, data mapping, common failure modes, and post-migration validation steps for teams moving from Freshdesk, Intercom, Help Scout, or other platforms.

Tejas Mondeeri Tejas Mondeeri · · 11 min read