OutcomeArc MCP Server

Your agent.
Your support manager.

OutcomeArc’s built-in AI handles individual customer questions. Bring your own agent to manage the bigger picture: what matters today, which customers are stuck, and what needs your attention.

Your agent. Your workflow. Only the access you grant.

Built-in AI for the tickets.
Your agent for the bigger picture.

You don’t need to bring an agent just to answer tickets. OutcomeArc already has built-in AI for customer conversations. Your own agent adds a support-management workflow: review the queue, investigate exceptions, and keep you informed in the tools you already use.

The Model Context Protocol (MCP) gives compatible agents access to that support context. You choose the inboxes and permissions; your agent uses the tools to answer your questions and carry out the work you delegate.

Manage support, not just tickets

Ask for the priorities.
Get the context to act.

Give your agent a support-manager brief, not a queue of replies to write.

Know what matters today

Ask for a brief across the inboxes you share. Your agent can review active and unassigned conversations, check who last replied, and explain which customers may need attention first.

Try asking your agent
What’s important today? Review our active conversations and give me the priorities, why they matter, and links to the threads.

Hear about the questions you need to answer

Have your agent investigate the thread, customer history, and available knowledge before asking for help. With scheduled checks and notifications configured in your agent platform, it can bring you the unresolved question and the context to decide.

Try asking your agent
Notify me when you find a customer who needs help with a question you can’t resolve. Tell me what you checked and what you need from me.

See the pattern behind the tickets

Look across conversations for recurring confusion, missing guidance, or product friction. Turn those findings into a short list of improvements for your team rather than treating every question in isolation.

Try asking your agent
What keeps coming up in support this week? Compare those questions with our knowledge and suggest what we should clarify or fix.
The toolset

Context to manage.
Tools to follow through.

Eight tools cover queue review, customer research, knowledge search, and replies when needed. Each connection sees only the tools its permissions allow. All requested permissions start checked during OAuth setup; uncheck any access you don’t want to grant.

See how access works
list_inboxes

Discover only the inboxes explicitly shared with this connection.

Any active connection
view_email

Read email or widget threads, including drafts and private notes, with authors, timestamps, status, and readable message text.

Read threads and customer profiles
search_emails

Search subjects, customer email addresses, and sent or received message text. Filter by status, assignment, team, or date; notes and drafts are not searched.

Search emails
recent_unassigned

Review recent active conversations without an individual assignee, even when a team is assigned. Results include assignment, last customer-message time, preview, and a conversation link.

Search emails
lookup_customer

Look up a customer by ID or exact email address and read their related conversation history within the specified allowed inbox.

Read threads and customer profiles
draft_email

Save a plain-text reply to an existing email conversation without sending or overwriting another draft.

Draft replies
send_email

Send a reply immediately when send permission is granted, with no OutcomeArc approval step. Confirmation is controlled by your client or agent harness. Delivery status is returned directly. To save without sending, use draft_email instead.

Send replies · High-risk write
search_knowledge

Search ready, enabled knowledge sources with citations. This is a separate permission for workspace-shared content.

Search workspace-shared knowledge
Read broadly within your granted inboxes. Write within clear limits. Read access includes drafts, private notes, widget transcripts, and inbox-scoped customer history. Writes are plain-text replies to existing email conversations, not new conversations or recipients, arbitrary CC/BCC, attachments, note creation, or widget replies. Your inbox signature and configured Auto-BCC still apply to sends.

Your agent can recommend priorities and improvements from the context it retrieves. The MCP tools do not change assignments, update knowledge sources, or manage outcome tracking.

Get connected

Bring your agent.
Keep your boundaries.

Use a remote MCP client that supports Streamable HTTP and OAuth with public-client registration and PKCE.

Remote MCP endpointhttps://app.outcomearc.com/api/mcp

OAuth recommended · No manually copied token needed

Client capabilities and setup screens vary. Use the URL above as the server address, not as a webpage to open.

  1. 01

    Add the server in your MCP client

    Choose a remote HTTP server and enter the endpoint. Start the client’s OAuth connection flow without a custom Authorization header.

  2. 02

    Sign in and choose access

    An OutcomeArc workspace owner or administrator selects the inboxes to share. All requested permissions—including knowledge search, drafting, and sending—start checked. Uncheck anything you don’t want, then authorize the connection.

  3. 03

    Give your agent its management brief

    Start with “What’s important today?” Tell it what matters to your team and when to ask for your input. For ongoing reviews, configure scheduled checks and notifications in your agent platform. Manage or revoke connections in Settings → Agent access.

    Open Agent access

Clients that omit OAuth scopes are offered all five permissions. An explicit scope request stays limited to those permissions. Existing read-only connections stay read-only; reauthorize to add sending or knowledge search.

Client doesn’t support OAuth?

If your client supports custom headers, an owner or administrator can choose Manual token fallback in Agent access, name the connection, and select its inboxes and permissions. The token is shown once and expires after 90 days.

Store it in your client’s secret storage and supply it as an Authorization: Bearer <token> header. Never put a token in a URL or share it in chat. Revoke the connection to rotate it.

Control stays with your team

Useful access.
Not an open door.

Choose inboxes explicitly

No all-inboxes wildcard. New inboxes are never added to an existing connection automatically, and permissions are checked on every action.

Choose whether your agent can send

Send permission starts checked when requested during OAuth setup; uncheck it for a connection that cannot send. Once granted, it sends immediately without an in-app approval queue. Sending is marked as a high-risk write; configure confirmation in your client or agent harness. OutcomeArc does not enforce a human confirmation step.

Revoke access when needed

OAuth access tokens last up to 30 days, capped by the connection’s expiry. Revoke a connection in settings to invalidate its access and refresh tokens. Revocation cannot recall mail already accepted for delivery.

Control knowledge access

Search knowledge is a separate checkbox, on by default when requested during OAuth setup. It covers the workspace’s shared knowledge—not just inbox-private content. Uncheck it to exclude knowledge search.

Also available · Experimental

Working in the browser?
Meet WebMCP.

In a supporting browser, WebMCP exposes the same scoped actions to a browser agent in your signed-in OutcomeArc tab. It’s an optional alternative to a remote MCP connection, not a requirement.

In Settings → Agent access, choose WebMCP, select inboxes and permissions, and choose Enable browser access.

  • Explicit opt-in for the current tab; no remote bearer token.
  • Use “Stop browser access” at any time. Reloading or signing out also stops access.
  • Browser grants expire within eight hours. Granted sends are direct, with confirmation controlled by your browser agent or harness—not an OutcomeArc approval queue.

WebMCP is experimental and not available in every browser. If unsupported, use remote MCP. These permissions constrain the exposed tools, not a browser agent’s separate ability to control your browser.

Start with one question:
What needs my attention?

Connect your agent to the support context it needs to answer.