LoginSign Up
← Back to Support
Weekly · Support · Week of July 27, 2026

Customer-support crosses from AI-assist to autonomous agents, with MCP as the default surface

agentic-aimcpassist-to-agentsrevenue-linked-supportmodel-churn
O
Generated 1h agoDrawn from 9 products

The week in customer-support

The week's throughline is unambiguous: support tooling is finishing its crossing from AI-assist to autonomous agents, and MCP has become the standard surface vendors build that around. Xurrent shipped the sector's clearest marker — Sera AI's first two named, autonomous ITSM agents, backed by knowledge-readiness checks and an Agent Designer — the pivot its prior months of Studio, MCP directory support, and CMDB hygiene were scaffolding toward. Formbricks supplied the other spark from the opposite direction: an OAuth-secured MCP server with first-class Claude and Codex setup guides, reframing a survey widget as an AI-accessible feedback platform.

Underneath the two sparks, nearly every active product is grinding in the same direction. Plain promoted its Sidekick agent to a first-class surface that acts on threads proactively; Canny crossed 55 MCP tools while wiring Autopilot into CRM pipeline; Richpanel kept folding the Shopify stack into an inbox its AI agent can act from; and LiveAgent quietly extended MCP knowledge-base tooling beneath a stability grind. The pattern is not new demos — it is the unglamorous plumbing (tags, CMDB, credit pools, connectors, OAuth) that makes agents operable in production.

Leaders

Xurrent is the week's headline. Sera AI moved from a tunable chat widget to two autonomous agents doing work inside the service desk, with Golden Set testing, readiness checks, and Agent Designer success metrics arriving in parallel. This is a vendor productionizing agentic support rather than previewing it.

Formbricks landed 5.2.0-rc.1, its biggest architectural turn: an OAuth-secured MCP server, a migrated auth stack (NextAuth to Better Auth), a Tailwind v4 rewrite, and a rebuilt Unify Feedback module with sentiment and emotion charts. In parallel it backported organization/workspace scoping across both the 5.1 and 5.2 trains — agent-accessibility and tenant-isolation shipping together.

Plain gave Sidekick its own full-page home and, more consequentially, had it start investigating, summarizing, and drafting the moment a thread matches a workflow's conditions. Workflows are becoming the orchestration layer that decides when the agent runs, moving Sidekick from something you talk to toward something that works your queue.

Canny overhauled its reporting tab into Ideas, Autopilot, and Portal sections, surfacing Autopilot as a first-class domain, while its MCP server crossed 55 tools and Autopilot began auto-linking Salesforce and HubSpot opportunities to captured feedback. The direction is feedback-driven revenue ops that agents, not just people, can operate.

Richpanel stayed on its integration tear, adding Rivo Loyalty & Referrals to a run that already absorbed Klaviyo SMS, RingCentral voice, and the AfterShip tracking/returns suite. Each connector widens what the AI agent can resolve autonomously from one inbox rather than adding a new core capability.

Wildcards

DoneDone is the off-pattern note: while the sector chases agents, it is quietly rebuilding itself around Kanban board by board — sort options, collapsible columns, and now drag-to-reorder checklist templates. Steady project-management craft with no AI thesis attached, a reminder that not every support-adjacent tool is racing to ship an agent.

Themes that compounded

  • MCP as the default integration surface: Formbricks shipped an OAuth-secured server, Canny crossed 55 tools, LiveAgent extended KB tooling, and Xurrent added directory PAT support — the standard is settling fast.
  • Assist features graduating to autonomous agents: Xurrent's Sera agents and Plain's proactive Sidekick both cross from suggestion to action this week.
  • Revenue-linked support: Canny ties feedback to CRM pipeline while Richpanel collapses the ecommerce stack into the inbox, both making support legible in commercial terms.
  • AI plumbing before AI headlines: tenant scoping in Formbricks, AI-Budgets and credit pools in LiveAgent, and region-by-region Zia rollout in ManageEngine ServiceDesk Plus show governance and metering landing ahead of the agents they support.
  • Frontier-model churn as table stakes: Hatz AI added Opus 5, Sonnet 5, Grok 4.5, and Gemini 3.6 Flash to its picker within days of each release.

Watch this week

Watch whether Xurrent's two Sera agents expand their roster and start reporting autonomous-resolution metrics through Agent Designer — that is the first sector data point on whether shipped agents actually close tickets. Watch Formbricks converge its 5.1.5 and 5.2.1 release-candidate churn into stable cuts, since that determines whether its MCP server is production-ready or still stabilizing. And watch Plain and Canny for the next action types their agents can reach across connected tools — both are widening what an agent is allowed to do on its own, and that guardrail expansion is where the near-term risk and value sit.