Originally published on integrationstars.com. Facts checked against Celigo's developer documentation on November 3, 2026.
Celigo now gives you two first-party ways to work with an integrator.io account outside the UI:
-
Platform MCP: a hosted MCP server, run by Celigo, that lets an AI client (Claude, ChatGPT, Cursor, VS Code and others) read, build, run and troubleshoot your account in plain language. Generally available since June 18, 2026. Endpoint
https://api.integrator.io/celigo-mcp(plus EU, AU and CA). Nothing to install. -
Celigo CLI (
@celigo/celigo-cli): every REST resource as aceligo <resource> <verb>command, with profiles, JSON/YAML/table/CSV output and--jqfiltering. Generally available since April 5, 2026; needs Node.js 22+; MIT licensed.
Both call the same REST API with the same permissions. Neither can do more than the signed-in user or the token could do through the API.
Which one to use
| You want to | Use |
|---|---|
| Ask questions and make one-off fixes in plain language | Platform MCP |
| Script repeatable changes or run checks in CI | Celigo CLI |
| Let a coding assistant build and troubleshoot from your machine | Celigo CLI (with Celigo skills loaded) |
| Call integrator.io from your own code or a runtime without Node.js | REST API |
A common pattern: diagnose in a chat client, then turn the fix into a reviewed CLI change.
Not to be confused with the Celigo MCP Server (the mcpServers resource you build to expose your own tools) or MCP Management (Celigo's gateway for governing MCP access). A tool you defined in an MCP Server can only do what you built it to do; Platform MCP can do anything your user or token can.
What an agent can change
list_* tools read, upsert_* tools create or update, and delete_resource deletes. Four details from Celigo's tool reference:
- Delete never blocks. Dependents come back as warnings, "but the delete always proceeds".
-
An update is a full replace.
upsert_*with an_idis a PUT of the whole document;patch_flowexists for single-field flow edits. -
Stored secrets are guarded. Since the October 28 update,
upsert_connectionkeeps a stored secret when the******mask is sent back. - Failed writes are not retried. A write that fails with a 5xx or network error may have run, so check before retrying.
Five guardrails before production
- Bind the session to a sandbox. OAuth sign-in picks one account and one environment; a bearer token is already bound to its account and environment.
- Use a scoped token. Celigo warns a full-access token "can read, change, and delete everything in your account". Custom-scoped tokens are limited to the resources you select.
-
Use CLI permission modes, and know their limit. Profiles run in
read,operateorfull(defaultfull). A mode "never adds to what the credential itself may do": it is a seatbelt, the token scope is the lock. Platform MCP's docs describe no equivalent modes. -
Ask for the plan before every write. In the CLI:
celigo push --dry-run. -
Read the audit log by source.
list_audit_log_entriesfilters by source, includingmcpandcli. A separate token per agent makes each change attributable.
Integrations in git: pull, review, push
npm install -g @celigo/celigo-cli
celigo login
# Pull one integration into a folder and put it in git
celigo pull --integration <integration-id> --dir ./acme --no-symlinks
cd acme
git init && git add . && git commit -m "baseline"
# After editing: check, plan, then write back
celigo status
celigo diff
celigo lint
celigo push --dry-run
celigo push
# Write the same tree into another environment
celigo promote --to sandbox
Safety rules worth knowing: lint runs offline against Celigo's published API schemas; push refuses any resource changed since your pull; push refuses connection documents that still hold the ****** mask; a tree is bound to the API host and token owner it was pulled with; promote needs full mode on the target.
Where we would start (our recommendation, not Celigo's)
- Read-only triage in chat, against a sandbox or with a read-scoped token.
- Operational actions (retries, error resolution) with confirmation. They change destination systems, so treat them as writes.
- Structural changes (new flows, mappings, deletes) only through pull, pull request and push.
Full guide, with the release timeline and FAQ: integrationstars.com. Sources: Celigo Developer Platform, Platform MCP, Platform MCP tools, Celigo CLI, profiles and permission modes, local tree, 2026 changelog, @celigo/celigo-cli on npm.
Top comments (0)