Connect Integrations to Parlé: Branded Cards and Custom MCP Tools

Integrations let the Parlé agent do things outside a conversation — search Confluence, look up a Jira ticket, call your own service — by connecting it to a system your organization already uses. You pick an integration from the gallery in Org settings → Integrations → Tools. Each card is a system Parlé knows how to connect to, with its server address and authentication method already filled in — you supply only what is genuinely yours: a credential, or an approval at the provider.

• Confluence is a branded card. Its server address (Atlassian’s own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card, so there is nothing to type before you approve access. See Connecting Confluence for what it can do and which scope to pick.
• Fathom is a branded card in the Meetings category, available at User scope only: each member connects their own Fathom account so the agent can search their recorded meetings and read transcripts, summaries and action items. Its server address (Fathom’s own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card. See Connecting Fathom for the shared-thread and recording-consent warning, and how disconnecting works.
• Fireflies is a branded card in the Meetings category, available at User scope only: each member connects their own account from My Integrations. Its server address (Fireflies’ own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card. Transcripts the agent reads in a project conversation are visible to everyone in that thread, so read the consent warning first. See Connecting Fireflies.
• Linear is a branded card. Its server address (Linear’s own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card, so there is nothing to type before you approve access. Unlike a read-only integration it can create and update issues and comments, so User scope is recommended. See Connecting Linear for what it can do and which scope to pick.
• Miro is a branded card in the Whiteboards category. Its server address (Miro’s own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card. The agent can read boards and also create or edit content on them, so read Connecting Miro for which scope to pick and what it can change.
• Notion is a branded card. Its server address (Notion’s own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card. The agent can search, read, create and update pages and databases, and Notion’s access can’t be narrowed to read-only, so User scope is recommended. See Connecting Notion.
• Otter is a branded card in the Meetings category. Its server address (Otter’s own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card. It is User scope only and read-only: each member connects their own Otter account from My Integrations, and the agent can search and read that member’s transcripts and summaries but never change anything. See Connecting Otter, including the warning about shared-thread visibility and recording consent.
• Productive is a branded card for agency management: projects, time, budgets and resourcing. Its server address (Productive’s own hosted MCP server) and authentication (OAuth 2.0) are fixed by the card. It requires Productive’s Ultimate plan and can change billing data, so User scope is recommended. See Connecting Productive.
• Custom MCP Tool is the card for everything else that runs an MCP server (an external service that speaks the Model Context Protocol) but has no card of its own. You give Parlé its address and how to authenticate, and it behaves exactly like any other integration from then on.

Setting one up has two levels, and both are deliberate:

1. Connect it once for your organization. Only Org Admins can do this, so project members cannot point the agent at arbitrary external services.
2. Enable it per project. A connected integration is available to nothing until a project switches it on — so an integration that reaches sensitive systems can be limited to the projects that should have it.

Parlé calls the other system from the server side, never from anyone’s browser, and the credentials you store are never sent to a client.

​

Chat apps

The Integrations tab has a Chat apps section above the Tools gallery described on this page. Microsoft Teams is live: ask Parlé from a Teams channel linked to a project, and see the linked channels and who may read answers. Slack and Google Chat appear as cards marked Coming soon; they cannot be connected yet.

​

Choosing who the credential belongs to

A card that supports more than one scope asks who this connection is for when you add it: Organization, Project, or User (each member connects their own account). The choice decides who authenticates, whose access the agent uses for that call, and who can end up reading the result — it’s the single most consequential decision in setting an integration up, so it has its own page. Read it before you connect anything that reaches content not everyone in your organization should see. Confluence, Linear, Miro, Notion and Productive offer all three. Fathom, Fireflies and Otter offer User only: a meeting recording belongs to the person who made it, so each member connects their own account and these can’t be shared organization-wide or project-wide. Custom MCP Tool offers Organization and Project. A card only ever offers the scopes it declares support for.

​

Part 1 — Connect an integration to your organization

​

What happens

You pick an integration from the gallery and fill in whatever it still needs. Before saving, Parlé checks the address is safe and that the other system actually answers — so a typo, a firewalled server, or an address that points somewhere it shouldn’t is caught now rather than mid-conversation. That check runs on every integration, including one whose address was filled in for you — a pre-filled address gets no special treatment.

​

What Parlé checks before saving

• The URL must be HTTPS. Plain http:// is rejected.
• The URL must not point to a private, internal, or loopback address — localhost, 10.x, 192.168.x, link-local, and cloud metadata addresses are all refused, including when a public-looking hostname resolves to one. This protects your cloud environment from being reached through the agent.
• The server must answer a quick connection check within a few seconds. If it does not, you will see “MCP server is not reachable” or “MCP server did not respond within the timeout”, and nothing is saved.

​

Part 2 — Authentication

Choose the option that matches what the provider gives you. Every branded card (Confluence, Fathom, Fireflies, Linear, Miro, Notion, Otter and Productive) always uses OAuth 2.0; Custom MCP Tool offers both.

​

Two OAuth states that need your attention

Parlé renews access automatically, but two states do not resolve themselves:

​

Part 3 — Enable an integration on a project

​

What happens

Connecting an integration at Organization or Project scope makes it available; it does nothing in any project until that project switches it on. A User scoped integration works a little differently — see My Integrations. Toggle it off at any time to remove it from that project. Everyone on the project sees the same list — project members and Project Admins can see which integrations are enabled (shown as Enabled / Disabled), but only an Org Admin can change the toggles.

​

Part 4 — Using an integration in a conversation

Once an integration is enabled, the agent discovers what it can do automatically — you don’t describe it to the agent. Ask for what you want in plain language (“what’s on the Q3 roadmap page?”) and the agent decides whether to use it. While a tool call is in flight the message shows a short activity line — an external tool can take tens of seconds. If an integration is unreachable or its credentials have stopped working, the agent is told the call failed and carries on with the rest of the turn — one broken integration costs a tool call, not the conversation. For a User scoped integration, whose credential answers the call depends on who’s asking — see Choosing a scope for what that means when the conversation is shared.

​

What gets recorded

Every call is written to your organization’s audit log against the user, the conversation and the tool. The record covers what happened, not what was said: the tool’s name and category, whether it succeeded, failed or was cancelled, how long it took, and the size of the request and response. The request and response payloads themselves are deliberately never stored. A tool’s arguments routinely contain your own documents, and ordinary writes to the audit log are rejected — an entry can’t be edited or deleted through Parlé or by an ordinary database write — so keeping payloads there would put a second copy of your content somewhere Parlé’s own deletion paths can’t reach.

​

Part 5 — Maintaining an integration

None of this depends on your plan. If your plan stops including integrations you can’t connect a new one, but everything below stays available — so an organization is never left holding a credential in Parlé’s vault that it can’t take back out.

​

Troubleshooting

Still stuck? Contact Parlé support with your organization name, the integration’s name, and (for Custom MCP Tool) the server URL you’re connecting.