Zendesk (`zendesk`)¶

The zendesk connector indexes support tickets and their comments, plus users and organizations. Use it to search support history by meaning across every ticket.

How MFS sees it¶

zendesk://acme/
├── tickets/
│   ├── records.jsonl      record_collection  (preset-indexed)
│   └── comments.jsonl     record_collection
├── users/records.jsonl
└── organizations/records.jsonl

The zendesk.tickets preset embeds ticket subject and description (status, priority, tags as metadata). Comments, users, and organizations enumerate but need [[objects]] rules to become searchable.

Credentials¶

Zendesk uses email + API token. The auth layer appends the literal /token suffix to the email automatically.

https://<subdomain>.zendesk.com → Admin Center → Apps and integrations → APIs → Zendesk API.
Toggle Token Access on.
Add API token → label it mfs → copy the value.

The token inherits the bound user's role permissions.

Configuration¶

subdomain = "acme"
email = "alice@acme.com"
api_token = "env:ZENDESK_API_TOKEN"
max_read_rows = 50000

Save the file as zendesk.toml, then probe and index:

mfs connector probe zendesk://acme --config ./zendesk.toml
mfs add zendesk://acme --config ./zendesk.toml

Sync and freshness¶

The connector re-reads the ticket, comment, user, and organization collections on sync. Deletions are caught by full_scan, and max_read_rows caps each resource path and marks recall partial when the cap is hit.

Search and browse¶

mfs search "billing dispute" zendesk://acme/tickets/records.jsonl
mfs search "refund policy" zendesk://acme/tickets/comments.jsonl
mfs cat zendesk://acme/tickets/records.jsonl --locator '{"id":12345}'

Pitfalls¶

Only ticket records are preset-indexed; add [[objects]] for searchable comments, users, or organizations.
Comments are fetched per ticket and can be expensive on large tenants.
max_read_rows caps each resource path.

Zendesk (zendesk)¶