Confluence Is Now a Macha Connector: AI Agents That Read and Write Your Docs

As of the April 9, 2026 release, Confluence is now a Macha connector. That means your AI agents can search your Confluence spaces, read the right page, and act on what they find — drafting a reply, adding the troubleshooting steps as an internal note on the ticket, or even writing a brand-new runbook back into Confluence after an issue is solved. Macha sits on top of the helpdesk you already use (Zendesk, Freshdesk, Gorgias, Front); Confluence becomes the knowledge-and-action layer underneath it. This isn't a static "upload your wiki" import. It's a live, two-way connection to the documentation your team is already maintaining.

This post covers exactly what shipped: the seven tools, how authentication and token-expiry monitoring work, the two workflows that ship with it, and the honest watch-outs before you turn it on.

What actually shipped

The Confluence connector exposes seven tools your agents can call. Three are read tools, three are write tools, and one navigates page hierarchy:

The split matters. Read tools let an agent use your documentation; write tools let it contribute back to it. And every write operation — create, update, comment — requires confirmation before it executes. An agent never silently edits your knowledge base.

Connecting it: API token + auto-created source

Authentication uses an Atlassian API token. You generate the token in your Atlassian account, paste it into Macha, and the connector goes live.

Two details make this more than a bare API hookup:

• Connecting Confluence auto-creates a data source — the same behavior as the Google Workspace and Notion connectors. You don't have to separately wire up Confluence as a knowledge source; the act of connecting registers it so your agents can draw on it as context.
• Token-expiry monitoring. You enter your Atlassian API token's expiration date, and Macha emails you a reminder 30 days before it expires, then automatically deactivates the connector on a 401 error if the token lapses. This is not a cosmetic nicety. Atlassian API tokens expire between 1 and 365 days, with a default lifespan of one year (Atlassian docs) — so a connection you set up today will silently break on a date you've long forgotten unless something is watching it. Macha watches it.

The two workflows that ship with it

Macha ships two pre-built use cases that show the read and write sides of the connector. Both are wired to your helpdesk — Zendesk in these examples — with Confluence underneath.

1. Search internal docs to resolve tickets (read)

This is the one most teams turn on first. Trigger: a ticket is created in Zendesk. The agent reads the ticket, searches Confluence across your spaces for the relevant documentation, reads the matching page, and adds the troubleshooting steps as an internal note on the ticket — so a human agent has the answer ready before they even open it.

A concrete pass from the shipped example: a customer writes in with "We're getting a 503 error when trying to access the reporting dashboard." The agent searches Confluence for 503 error reporting dashboard, finds the "Reporting Dashboard — Troubleshooting Guide" in the Engineering space, and posts an internal note:

The 503 error on the reporting dashboard is typically caused by a cache overflow during peak hours. Steps: check cache service status, restart the reporting worker, verify the database connection pool. The guide notes this was last seen in January and resolved by restarting the reporting worker.

Nothing customer-facing happened automatically — it's an internal note, the safe default for a knowledge lookup. The human still owns the reply; they just don't have to go digging. (See the full Search internal docs to resolve tickets use case.)

2. Auto-document resolved issues as runbooks (write)

The write side closes the loop. Trigger: a ticket is solved. The agent pulls the full conversation, searches Confluence first to check a runbook doesn't already exist, and if none does, creates a new page documenting the problem, root cause, solution, and prevention — in your Support Runbooks space.

From the shipped example: ticket #4821, "SSO login failing for new users," gets resolved over six comments. The agent confirms no existing runbook, then creates "Runbook: SSO Login Failing for New Users" capturing the root cause (a missing department attribute in the IdP's SAML mapping), the fix, and a prevention note added to the SSO setup checklist.

Because Create Page is a write tool, that page creation asks for confirmation before it commits. Over a few weeks, this is how a support queue quietly turns itself into documentation instead of letting hard-won fixes evaporate when the ticket closes. (See Auto-document resolved issues as runbooks.)

Why a live connector beats a one-time import

The common way to "use Confluence with AI support" is to export or sync your wiki into a vendor's index once and answer from that copy. That approach has two well-known failure modes: the copy drifts out of date the moment someone edits the original, and it's strictly read-only — the AI can answer from your docs but can never improve them.

Macha's Confluence connector is different on both counts. It queries Confluence live at the moment the agent needs an answer, so you're always reading the current page, not a stale snapshot. And it's bidirectional — the same connector that reads your runbooks can write new ones. The top informational guides on this topic (eesel's Confluence + Zendesk guide among them) stop at read-only Q&A and skip the operational reality of token lifecycle entirely; the runbook-creation half of the loop and the 30-day expiry reminder are the parts that actually keep a deployment alive in month three.

To be clear about positioning: Macha is not a replacement for Confluence or for your helpdesk. It's the agent layer that connects them. Confluence stays your source of truth; Zendesk (or Freshdesk, Gorgias, Front) stays your ticketing system. Macha is the thing that reads one and acts in the other.

Watch-outs — when not to flip every switch

• Write tools deserve a slow rollout. Start with read-only workflows (search and internal notes). Turn on Create/Update Page only once you trust the agent's judgment, and lean on the built-in confirmation step rather than auto-approving writes. A confidently-wrong runbook is worse than no runbook.
• Search quality follows documentation quality. If your Confluence spaces are a graveyard of half-finished pages and duplicate guides, the agent will surface the same mess faster. The connector amplifies the knowledge base you have; it doesn't fix it.
• Scope your API token. Atlassian supports scoped API tokens that restrict a token to only the actions it needs (Atlassian KB). Give Macha's token the narrowest scope that covers the spaces and operations you actually want it touching.
• Mind the expiry date you enter. The 30-day reminder only fires if you enter the correct expiration date when connecting. Get it right, or the auto-deactivate-on-401 safety net is the only thing left between you and a silently dead connector.
• It counts as an integration. Connectors draw against your plan's integration limit, so a Starter team weighing Confluence against another tool should check the pricing page first.

What it costs

Like every action in Macha, Confluence tool calls run on credits — charged per AI action, not per resolved ticket. A typical search-and-note flow is a handful of actions; pricing per action depends on the model you pick (roughly 0.5–9 credits, with the default GPT-5.4 Mini at 1 credit). The Confluence connector itself counts toward your plan's integration allowance. There are no Confluence-specific add-on fees — see the pricing page for current plans, and note that Confluence Cloud's own Standard tier runs around $5.75/user/month (Premium ~$11), per public pricing at the time of writing (approximate — check Atlassian for current numbers).

FAQ

What can Macha do in Confluence? Seven tools: search pages, read a page, list spaces, get child pages, create a page, update a page, and add a comment. Reading is automatic; the three write actions (create, update, comment) require confirmation before they run.

Does Macha edit my Confluence without asking? No. Every write operation pauses for explicit confirmation. Read-only tools (search, get, list) run freely; nothing is created or changed without a yes.

Do I have to import my wiki into Macha? No. Connecting Confluence auto-creates a data source and queries your pages live, so agents always read the current version — there's no separate upload or sync step, and nothing goes stale.

How does authentication work, and what happens when my token expires? You connect with an Atlassian API token. You enter its expiration date, and Macha emails a reminder 30 days before expiry, then auto-deactivates the connector on a 401 if the token lapses — so a dead token never silently breaks your agents.

Is Macha replacing Confluence or my helpdesk? Neither. Macha is the AI agent layer on top of your existing helpdesk; Confluence remains your documentation source of truth. Macha connects the two.

Which plans include the Confluence connector? It's available as a standard connector and counts toward your plan's integration limit. See the pricing page for current allowances.

Try it

If your team's answers live in Confluence and your tickets live in Zendesk, this connector is the bridge that's been missing. Start a 7-day free trial, no credit card required, connect Confluence and your helpdesk, and turn on the Search internal docs to resolve tickets workflow first — it's the fastest way to feel the difference. Full setup details are in the Confluence integration page and the docs.

Written by Abbas (Customer Support & AI, Macha) · Reviewed by Ankeet Guha (Co-founder & CTO) · Published 2026-06-24 · Last updated 2026-06-24.