Changelog

September 18, 2026: MCP and Data ingestion

🧩 MCP server upstream connections

Connect a Celigo MCP server to third-party native MCP servers such as Salesforce, HubSpot, and Snowflake, and govern all of their capabilities from one place. Attach an upstream MCP connection to bring in its tools, resources, and prompts, which are discovered automatically and can be enabled or disabled individually. Override the name, description, and annotations on a proxied capability so AI agents see your organization's guidance rather than the generic upstream text, and add Celigo-built capabilities alongside proxied ones on the same server. Your existing governance applies to every call: authentication, end-user groups, capability sets, and request logging. Existing MCP servers are unaffected. Learn more about the MCP server proxy.

🙋 MCP server end-user connections

End users authenticate to downstream apps with their own credentials. A connection can be marked as requiring end-user credentials, and child connections store only the end user's credentials while referencing the parent at runtime. Learn more about MCP server end user connections.

📥Data ingestion - NetSuite JDBC as a Celigo Sync source

Connect NetSuite JDBC as a native source application in the Celigo Sync wizard and select multiple NetSuite objects for a pipeline in a single pass, rather than hand-writing a separate query per object. For detailed information on sync, see Create a sync from NetSuite JDBC.

📥Data ingestion - Microsoft SQL Server as a Celigo Sync destination

Load data through Celigo Sync into Microsoft SQL Server, which joins Snowflake and NetSuite Analytics Warehouse as a destination for sync pipelines. For detailed information on the sync, see Create a sync to Microsoft SQL.

📥Data ingestion - Object-level selection for manual runs and full resyncs 

Choose which objects to include when you start a manual run or a full resync on a sync, instead of running every object each time. Resync a single object after adding a column at the source, or re-run only the objects that failed. The selection list shows each object's last-run status, and objects you don't select keep their data, sync state, and last-run status untouched. Run details reflect only what you selected, so record counts, durations, and errors match the actual run. Running with all objects selected behaves as it does today, and scheduled runs continue to process every object on the sync. Learn how to choose objects to include in a sync.

September 16, 2026: SFTP file server, Celigo Topics, SCIM, MCP rate limits, network policies, and more

📁 Celigo-hosted SFTP file server

Give every trading partner a secure SFTP endpoint hosted by Celigo, with no servers to run on either side. Enable a hosted file server per environment, then create an isolated account per partner with its own password and a Celigo Storage folder as its locked home directory. Partners connect with any standard SFTP client on port 22, and the files they upload become ordinary Celigo Storage files the moment they arrive — visible in the Storage UI, governed by storage usage, and consumable by flows through the Celigo APIs connector. Learn more about the Celigo-hosted file server.

📡 Celigo Topics

🪪 SCIM provisioning

Automate user lifecycle management from your identity provider. Provision and deprovision the people who build and operate integrations from a new SCIM Provisioning section in the account's Security and SSO settings, and manage end users from a matching section in each external IdP configuration. Learn more about provisioning users with SCIM.

🚦MCP server rate limits and usage quotas

Configure three optional traffic limits on each MCP server. A rate limit caps the total tool calls the server accepts in a time interval, protecting downstream systems from bursts. A per-user rate limit gives every end user their own allowance, so a single runaway AI agent can't consume the server's capacity and degrade service for everyone else. A usage quota caps cumulative tool calls over a longer period. When a limit is reached, requests receive a standard 429 response telling the client when to retry, and throttled requests appear in the MCP server logs with the consumer's identity. All three are optional — leave them blank for unlimited access.  Learn more about MCP server rate limits and usage.

Intelligent document processing for AI agents

AI agents extract structured data from invoices, purchase orders, and Excel files into a schema you define, with field-level confidence signals. Agents detect document extraction intent and invoke processing automatically, with no setup beyond the existing agent schema. Learn how to extract data from documents with an AI agent.

🧱Network policies

Administrators can now define reusable IPv4/IPv6 allow and deny lists under Settings > Security and apply them across workspace logins, MCP servers, end-user authentication, and API tokens without re-entering ranges (PATs inherit workspace policies automatically). Deny lists take precedence, empty allow lists leave access unrestricted, and enforcement only begins once a CIDR is configured, ensuring existing endpoints experience no immediate behavior changes. Blocked logins and calls are recorded in audit and request logs, and full CRUD API support is available for programmatic management.  Learn more about network policies.

📑EDI support for file provider connectors

Parse and generate EDI X12 and EDIFACT documents on file provider connectors, including Google Drive, Box, Dropbox, Azure Blob Storage, Google Cloud Storage, and Celigo Storage. EDI profiles were previously available only on FTP, S3, AS2, and VAN, so EDI flows had to route through one of those systems regardless of where the trading partner actually exchanged files. Choose EDI X12 or EDIFACT as the file type on a file provider export, import, or lookup, then select an EDI profile exactly as you would on an FTP or S3 flow step. Parsing and generator definitions, processing of parsed data, and functional acknowledgement generation behave the same as on FTP and S3, and Celigo Storage now handles EDI, so you can stage EDI documents between flows without external file infrastructure. Configure it through the UI or the v1 API. Existing EDI flows on FTP, S3, AS2, and VAN are unchanged. Read about the EDI connectors for B2B manager.

🔐Tenant-wide SSO enforcement

A new account-level setting requires every user to sign in through the identity provider, so no one is left on optional SSO. Account owners can now require SSO for themselves as well, which was previously impossible and removes the last username and password path into the account; that change is confirmed with a live SSO re-authentication first to protect against lockout. If the identity provider becomes unavailable, access is restored through a support-assisted, identity-verified recovery process. Both settings are off by default. Learn more about requiring SSO for the account owner and all account users.

☑️ Object-level selection for manual runs and full resyncs 

Choose which objects to include when you start a manual run or a full resync on a sync, instead of running every object each time. Resync a single object after adding a column at the source, or re-run only the objects that failed. The selection list shows each object's last-run status, and objects you don't select keep their data, sync state, and last-run status untouched. Run details reflect only what you selected, so record counts, durations, and errors match the actual run. Running with all objects selected behaves as it does today, and scheduled runs continue to process every object on the sync. Learn how to choose objects to include in a sync.

💾 Shift JIS file encoding on FTP and file connectors

Choose Shift JIS as the file encoding when you export, look up, import, or transfer files on FTP, SFTP, FTPS, and other file connectors. Japanese files encoded in Shift JIS show the correct characters in preview and in your flow, and files you write back use the encoding the destination system expects. See File transfers. 

September 16, 2026: Ora

✨ Ora

This week, Ora gained ground in trading partner connections, account-wide visibility, and AI agent building. It can now set up an EDI connection directly from Celigo's published connector templates, report flow trends across an entire account instead of one integration at a time, and build a working AI agent and the flow that runs it from a single plain-language request. MCP server management also picked up two new hands-off actions for connections and end users. 

Ora now runs a real Playground agent that runs templates, scripts, and parsers against your pasted data instead of just describing it. Ora also gets more self-sufficient with connections: it can swap a connection across an entire flow, clear the way for a stuck deregistration, and act on plain statements like "the timeout should be 30 seconds" instead of just describing the current setting. Separately, Ora's scripting knowledge now covers the platform's own lookup-cache and export/import APIs; it can tell you exactly where flow notes and TODOs belong, and every new chat now gets a meaningful title automatically.

Connect to EDI trading partners

Set up EDI connections from published connector templates: Ask Ora to connect to a trading partner such as Target over AS2, and it walks the same Protocol → Region → Business model path as the connection form's Trading Partners picker, filling in the partner's template values and asking you only for the details only you can supply, such as your SFTP host or AS2 identifier. If Celigo hasn't published a connector for that partner, Ora points you to the request process instead of building a placeholder connection.

Diagnose and fix errors

Account-wide flow trend reporting: Ask for flow error trends, success rates, or processing time across your whole account, and Ora returns a real day-by-day series for every flow instead of a rough snapshot. On large accounts, it asks you to narrow the request rather than stalling.

Visibility into legacy two-step flows: Ora can now see the steps inside legacy two-step flows, including most 1.0 Integration App flows and Data Loader flows. These used to look empty to Ora, so step-level fixes such as mapping edits now work on them too.

Live fact-checking in connected apps: When the evidence doesn't settle what caused an error, Ora reads the live record in the connected external system, such as NetSuite, to confirm what's actually there instead of leaving that check to you.

MCP servers and AI agents

Per-server connection overrides: Point a published Tool at a different connection for a specific MCP server — for example, use a sandbox connection on your QA server — directly from chat.

Sign-in provider visibility: Ask Ora to show your account's external sign-in providers, such as Okta or Auth0.

Suspend and re-enable MCP end users from chat: Tell Ora to disable or re-enable an end user's MCP access, or remove a person's direct MCP server grant, and it stages the change as an approval card without a trip to the End users page.

Build AI agents from real integration patterns: Describe what you want an agent to do — read a mailbox and its attachments, classify documents from storage, answer questions in a Slack thread, or wait for your approval before acting — and Ora builds it using the same patterns behind the Marketplace's AI agent templates.

Agent requests that come with their own flow: When a request needs a flow around the agent, such as "an agent that watches X and suggests Y," Ora asks where the input comes from and where results should go, builds the flow with the agent inside, and hands you the exact steps to take it live.

Talk to Ora your way

Remembering your preferred level of detail: Ora can now offer to remember how much explanation you like and carry that preference into future chats.

A live sandbox for testing expressions

Playground agent: On the Playground page, Ora now renders, diagnoses, and writes Handlebars and SQL templates, filters, transform rules, and scripts, and parses CSV and XML, against the actual data you paste in, run through the same processors the editors use. Paste an expression and sample data from any page, and Ora routes it to an agent that can run it for real instead of just describing what it should do. Nothing on the Playground is saved; when you want the result applied to a real export or import, Ora hands it off and stages the change for your approval.

Connections that manage themselves

Connection swapping: Ask Ora to replace one connection with another, and it performs the swap itself across an entire flow, including API and Tool steps, staged for your approval, instead of walking you through the menus by hand.
Clearing the way to deregister a connection: When a connection won't deregister, Ora now shows exactly what's still using it, such as flows, API or Tool steps, aliases, or syncs, proposes moving everything to a same-type replacement, and deregisters the connection once you approve.
Acting on "should be" statements: Tell Ora "the timeout should be 30 seconds" on a connection, and it now treats that as a change to make, staged for approval like any other edit, instead of just describing the connection's current setting back to you.
Per-instance webhook URLs: For a multi-instance flow, Ora can now generate a separate, working webhook URL for a single instance by cloning the template's listener, so a webhook hit reaches only that instance.

Scripts and knowledge that go deeper

Scripting against the platform's own APIs: Ora now knows Celigo's built-in script API. Ask it to read or update a lookup cache, keep state across script runs, or run an export or import from inside a script hook, and it writes working code instead of saying it's impossible.
Guidance on where notes and planned work belong: Ask Ora whether you can add a comment or TODO placeholder to your flow, and it points you to the right spot, such as a flow or step description or the integration readme, and offers to write the note for you.
Integration App edition awareness: Ora can now see the installed edition of an Integration App against your licensed edition, so a question like "why is this setting missing?" gets a direct answer, such as which edition includes it and how to upgrade, instead of a hedge.

Keep your chat history organized

Automatic chat titles: Every new Ora chat now gets a short, meaningful title generated automatically from the first message, including chats that start with just a file upload. A vague opener, such as a greeting or a lone file, gets its title upgraded after Ora's first reply so your chat history stays easy to scan. Renames you make by hand always stick.

September 4, 2026: Ora

✨ Ora

This week, Ora expands what admins can do with MCP end users, sees deeper into API and Tool run history, and works more fluidly across multi-instance flows and multi-environment accounts. It also speaks the platform's newest Handlebars additions and recognizes three new file providers, with clearer file-metadata handling across the board.

Know and manage who's using your MCP servers

• See who's using your MCP servers: Ask "who are my end users," "did Jane accept her invite," or "who's in the Finance group," and Ora answers from live rosters and group membership across the account.
• Understand exactly what access someone has: Ask what a specific end user or MCP server grants, and Ora shows which direct, group, or role grant delivers each permission, and flags access that's granted but unusable.
• Find and trace end users from any page: Search the account for an end user or group by name or email and see their groups, servers, and invite status. Ask what would be affected if you deleted an MCP server, and Ora lists every end user and group that would lose access. This data stays visible to admins only.
• Invite, revoke, and manage end users and groups from chat: Ask Ora to invite or revoke an end user, or create, update, and manage end-user groups and membership. Every change stages as an approval card first, and the audit log attributes it to Ora.
• Grant MCP server access by group: Ask Ora to give a group access to an MCP server, and it stages the grant with a before-and-after summary.

Review your history

• Ask about API and Tool run history: Ask Ora to show an API's failed requests from last week, its busiest hours, or when a specific record came through. Ora walks a single request step by step with the exact payloads, and can turn log capture on or off or purge logs on request. The same capability extends to Tools, and activates automatically once the platform's Tool invocation history feature is live.
• See exactly what Ora has changed: Ask "what has Ora changed recently," and Ora now queries the audit log using the platform's dedicated attribution source, so its own changes are correctly identified.

Work across flows and environments

• Work with multi-instance flows: Ask about a multi-instance (template) flow, and Ora lists every instance and which are enabled. Ask Ora to run every instance of a flow, and it starts each enabled instance, skips paused ones and the template itself, and reports what actually ran.
• Work across production and sandbox: Ora now knows that environments are separate accounts under the hood. Point it at a resource in another environment, and it explains the boundary instead of guessing wrong. Ask it to copy your sandbox mappings to production, and Ora writes a ready-to-paste handoff message for the Ora session in that environment, which does the work and replies with an answer to paste back.
• Trace exactly where a flow's data stops: Ask why data isn't reaching a step, and Ora runs the flow in test mode and narrates the record's journey, including which branch it took, which filter dropped it, and where it stalled, even on flows that have never run. Ora can show the exact request a step would send without leaving the building, or, only when you say so, actually send it and show the real response.

Write smarter templates and imports

• Author date math with readable units: Ask Ora to add three days to a date, and it produces {{dateAdd date 3 "d"}} with readable interval units instead of raw milliseconds.
• Read a lookup cache directly in a template: Templates can now pull a value straight from a lookup cache with {{getLookupCacheData}}, and the template editor validates the result against your real integration and flow context.
• Fan out AI agent and guardrail imports: Ora now understands one-to-many fan-out for AI agent and guardrail imports, where a single incoming record with an array becomes one agent invocation per child element.

Handle more files, more precisely

• Move files from more providers: Ora now treats Google Cloud Storage, ADLS Gen2, and Celigo APIs as full file providers. Ask it to move files from a GCS bucket, and it builds the correct folder-transfer flow, including GCS's own bucket-and-key-prefix addressing.
• Stamp file provenance on more exports: File-reading exports and lookups across FTP, S3, AS2, local file system, NetSuite file cabinet, and cloud-storage folders can now stamp each record with the source file's name, size, and last-modified date.
• Get a clear answer when a stamp isn't possible: If you ask for a file-provenance stamp on a step that doesn't read files, Ora explains why that can't work and offers a record-field alternative instead of quietly saving something broken.
• Retarget file exports by folder: Ora can now point a file-provider export at a new folder ID directly, updating the real directory binding rather than the display path alone.

September 2, 2026: Ora

✨ Celigo Ora: Beta through GA

From the March 2026 beta to general availability, new Celigo Ora capabilities brought major enhancements to the integrator.io platform. You can now leverage flexible interaction options, deeper account-specific context, dynamic error diagnosis, and tighter change guardrails. Additionally, these features deliver expanded search, approval, and audit capabilities across the entire environment. The following sections outline what shipped by capability area.

Talk to Ora your way

• Run multiple conversations at once: Open Ora in several browser tabs, or start several conversations in one tab, and keep separate pieces of work moving in parallel.
• Dictate instead of typing: Speak your prompt, and Ora transcribes it, including long dictations, in your language.
• Attach files and screenshots: Drop an error screenshot, a CSV, or an API spec into the chat, and have Ora work from it. You can add references to Celigo Storage files in your chats.
• Ora inside the mapping and expression editors: Mapping and expression editors now have Ora built in, replacing the legacy AI helpers. Ask "what does this field do?" for an explanation, or ask for a change and review it as a diff before it's applied.

What Ora knows about your account

• Teach Ora once, and it remembers: Say "remember this," and Ora saves it either in the integration's readme for team knowledge or as a personal rule that follows you across every account you work in. Every saved instruction is a visible, editable file, not a hidden memory store.
• Integration App expertise: For managed apps such as Shopify–NetSuite and Amazon–NetSuite, Ora knows what each flow does, which setting governs each behavior, and the documented fixes, grounded in official Celigo docs and your account's configuration.
• Custom integration awareness: Ora builds a working profile of each custom integration, including what it does, which pieces are risky to change, and its normal volumes, and uses that profile to answer questions, diagnose errors, and search your account.
• Flexible account search: Search by name, rough description, configuration (for example, "all disabled flows"), relationship ("which flows use this connection"), or count ("how many flows per integration"), and get complete, exact answers instead of a top-10 sample.
• Contextual awareness: Ora knows what you're working on and recommends prompts to help you build and troubleshoot quickly.

Diagnose and fix errors

• Evidence-based diagnosis: Ora pulls the actual failed record and request/response data, checks whether a recent configuration edit lines up with when errors started, and states what's confirmed versus what's a hypothesis, plus the one check that would settle it. Known Integration App errors surface their documented fixes.
• Live fact-checking in connected apps: Ora can read the live record in NetSuite, Salesforce, or your database (read-only) through connections you already have to confirm details such as whether a record still exists or is still open.
• Bulk error actions by record content: Describe a group of errors by what's inside the failed payload, such as orders with no line items or records where the country isn't US, preview the count, and act on the whole set.
• Plain-language retry fixes: Describe a correction, and Ora applies it across one error or thousands, using the platform's own processor so records aren't reshaped or dropped. Every fix stages for your approval.
• Record tracing and replay: Trace a record through a run step by step, or replay a failed record through a step's filters and mappings without sending anything to the destination. Ora can also explain why a record was ignored, check whether a run is stuck, read connection debug logs, and inspect lookup cache contents, all from chat.
• Anomaly detection over any time window: Ask about anomalies over any date range, for example "any anomalies in my flow executions in the last 15 days?", and Ora checks every day in that window, catching error spikes, volume drops, and flows that quietly stopped running on the day it happened. Previously, Ora could only assess the last 24 hours.Builders running high-volume NetSuite integrations or data warehousing workflows, and developers who prefer REST over SOAP or need direct access to custom NetSuite record types.

How Ora keeps changes safe

• Self-testing before saving: Every mapping, template, query, filter, and script Ora writes runs through Celigo's real processors against your sample data, is judged against what you asked for, and is corrected until it's right. If something can't work, Ora says so instead of saving a broken artifact.
• Staged approvals with verification: Every change stages as an approval card, validated against the platform before you accept it, individually or all at once. Ora tracks what you approved, resends any card that didn't land, verifies after acceptance that the change persisted, and attributes every action in your audit log.
• Builds that finish the job: Flow, API, and Tool builds are grounded in real sample data at every step, verified end to end, and auto-repaired before delivery. You can keep editing a flow while Ora builds it, since it merges your changes rather than overwriting them. Quick edits such as renames, schedules, and enable/disable now stage in seconds.

Find and install from the Marketplace

• Search by meaning and template content: Describe your problem in your own words to find the right Marketplace listing, search inside every template for the specific flow or agent tool you need, and check which candidates actually fit your account before you install.
• Marketplace browsing and install visibility: Browse and install AI Agent listings directly from chat, check what's new in the Marketplace, and see upfront during a wizard install which external steps are already complete in your account.

Expanded platform coverage

• Real-time database listeners: Build change-data-capture exports from plain language for MongoDB change streams, SQL Server CDC, and PostgreSQL logical replication.
• MCP servers and AI agents: Build MCP servers complete with tools and prompts, and publish Celigo Storage documents as read-only MCP resources. You can also create Claude-powered AI agents with your own Anthropic key, subscribe agents to governed MCP resources, and diagnose errors on AI agent flow steps.
• Integration lifecycle management for APIs and Tools: Snapshot, pull, and revert now cover APIs and Tools in addition to flows, with previews and field-level diffs before anything merges, plus lifecycle-linked clones that stay connected to their source.
• Deeper configuration coverage: Author custom settings forms with live dynamic dropdowns pulled from NetSuite, Salesforce, and your saved searches; manage Integration App settings (1.0 and 2.0); use Handlebars on object and array fields in Mapper 2.0 and directly in filter and branch rules; add JavaScript hooks, filters, transforms, and branching on flows, APIs, and Tools; and create Personal Access Tokens directly from chat.

Celigo Ora overview

Error management with Celigo Ora

September 2, 2026: Celigo Storage, AI agent models

📁 Celigo Storage integration folders and token path scoping

Each integration now gets its own system folder, with permissions inherited from your existing access control, so Manage All, Monitor All, and Custom-role users reach Storage through their integrations rather than needing account-wide access. API tokens can be scoped to specific folder paths, limiting what a given credential can read or write.

Celigo Storage overview

Work with files and folders in integrator.io

Restrict an API token to specific Celigo Storage paths

Four more model providers for AI agents and guardrails

• Mistral, xAI, Hugging Face, and DeepSeek join OpenAI, Gemini, and Anthropic in the model picker — BYOK via your own provider connection, and AI Credits are not charged for BYOK runs
• Agent and guardrail settings adapt to the selected model — you only see the controls that model supports
• Existing agents are untouched, and invalid provider/model/option combos are rejected on save with an error that names the problem

August 31, 2026: NetSuite REST API support, Anthropic in Celigo AI

🛠️ NetSuite REST API Support (SuiteTalk + Custom RESTlets)

The NetSuite connector now supports REST APIs as a first-class API type for record-based exports, lookups, and imports — covering both SuiteTalk REST Web Services and Custom RESTlets, directly within the NetSuite connector without requiring a separate generic HTTP connection.

Why it matters:

NetSuite’s Saved Search — the basis of most existing NetSuite exports — is known to be inefficient at high volumes. SuiteTalk REST Web Services provide a faster, simpler alternative for large-volume and data warehousing workflows, and don’t require the additional SuiteAnalytics Connect license that the JDBC approach requires. REST also gives builders simpler access to custom records and fields. With this addition, Celigo now supports all three major ways to integrate with NetSuite: Celigo RESTlets, SOAP Web Services, and SuiteTalk REST.

Who it's for:

Builders running high-volume NetSuite integrations or data warehousing workflows, and developers who prefer REST over SOAP or need direct access to custom NetSuite record types.

What changed:

New API type in the NetSuite connector. A REST APIs option is now available in the “How would you like the records exported/imported?” selector on record-based NetSuite export, lookup, and import steps. Selecting REST APIs exposes two sub-options: SuiteTalk (configurable via a NetSuite-oriented Simple view or an HTTP view for advanced control) and Custom RESTlets (configurable via HTTP view). The platform automatically sets the appropriate NetSuite REST base URI based on the selection. File-based flows are unaffected. Existing Celigo RESTlets, SOAP Web Services, and generic HTTP-based integrations continue to work unchanged.

Import NetSuite data

Export NetSuite data

🛠️ Anthropic (Claude) in Celigo AI [Improved]

You can now run AI agents on Claude models without bringing your own Anthropic API key. 

Why it matters:

Until now, using Claude in Celigo meant BYOK — your own Anthropic account, key management, and separate billing. That's gone: Claude is now hosted under Celigo AI, selectable just like OpenAI and Gemini, with usage billed through Celigo.

Who it's for:

Anyone building with AI agents, guardrails, or AI chat — especially teams without their own Anthropic agreement 

What changed:

• Claude models (Opus, Sonnet, and Haiku) are now selectable under the Celigo AI source for AI agents, guardrails, and chat.
• Full parity with OpenAI and Gemini — tool calls (IO tools and MCP servers), structured outputs, streaming, extended thinking, and Anthropic web search.
• BYOK is unchanged — existing Anthropic, OpenAI, and Gemini setups keep working as-is.

Learn about Celigo AI agents