Bringing your full OTT live video streaming workflow into the AI tools your NOC team already uses
B2B
OTT/Streaming Infrastructure
Authenticated MCP Server
Touchstream’s customers need their operations teams to get to the root cause faster, before viewers are impacted. By building an authenticated MCP server with Alpic, OTT streaming observability data is now accessible directly inside Claude, Cursor, or any MCP-compatible AI client, securely, privately, and in plain language.
Touchstream is the OTT operations hub trusted by some of the world's largest broadcasters and streaming platforms including Sky, DAZN, Comcast, Canal+, and NBC Peacock. Their VirtualNOC platform gives NOC teams end-to-end visibility across the full video delivery chain: encoder, origin, CDN, SNMP, and QoE, all in one place, correlating signals across every vendor in the stack so that the answer to any incident is available immediately.
When something breaks during a live stream, every second counts. Touchstream gives operations teams the data they need to diagnose fast, escalate smart, and restore service before viewers notice. The question was how to make that intelligence even more accessible, especially as AI tools were becoming a daily part of how engineering and operations teams work.
The challenge
Touchstream's platform generates rich, real-time observability data across every layer of the OTT stack. But accessing that data meant logging into a dashboard, navigating to the right view, and manually correlating signals across CDN, encoder, and quality metrics.
For NOC engineers working fast under pressure, that friction matters. And as AI clients like Claude and coding agents became standard tools in enterprise technical workflows, there was a clear opportunity: bring the data to where the team already is, rather than requiring them to go somewhere else.
The challenge was doing this securely. Touchstream's customers operate in multi-tenant environments where each team should only ever see their own data. Any AI-native interface had to respect that boundary from the start.
The solution
Alpic designed and deployed a production-grade authenticated MCP server for Touchstream, built in TypeScript on the Skybridge framework and hosted on the Alpic platform.
The server connects Touchstream's video operations data to any MCP-compatible AI client. Engineers can now query their live stream health, investigate incidents, and generate handover notes through natural language, without leaving the tool they're already working in.
Authentication is handled via OAuth 2.0 pass-through. Rather than the MCP server holding a single set of API credentials shared across all users, each user authenticates with their own Touchstream account when they connect through their AI client. The server forwards that identity to Touchstream's backend on every request, so access controls are enforced exactly as they would be in the Touchstream dashboard. If a user only has access to three streams in the platform, they only see those three streams through the MCP server. There is no shared context between tenants, no way to query another customer's data, and no anonymous access.
The servers supports four operational intents, built around the most important workflows that Touchstream enables:
What's wrong right now
See what's broken, which streams are affected, and where in the stack the fault sits across CDN, origin, headend, and QoE. Turn a wall of alerts into a clear starting point for triage.
Incident playback
Answer "what actually happened?" in hours, not days. For a given incident, replay the T-30 to T+30 window and get a structured summary of what changed across encoder, origin, CDN, SNMP, and quality of service. No more manual log correlation.
CDN and region comparison
Rank CDNs or regions by error rate, latency, throughput change, and quality deltas. Make real-time decisions from shared, independent data during a live event or post-incident review.
Handover and ticket drafting
Generate structured incident notes, escalation payloads, and executive or customer-facing summaries directly from the data: consistent outputs, ready to paste into a ticket or send to a stakeholder, regardless of who is on shift.
What Alpic built
Authenticated by design
The MCP server uses OAuth 2.0 pass-through from day one. No user ever sees another tenant's data. This wasn't an afterthought, it was a core architectural requirement baked in before the first tool was written.
Multi-tenant from the ground up
Isolation is enforced at the software level across all tenants. Each customer operates in their own context, regardless of which AI client they use to access the server.
Focused tool surface
Rather than exposing every API endpoint, Alpic designed a small, precise set of tools mapped to the operational scenarios that matter most. Fewer tools means less ambiguity for the model and more reliable outputs for the engineer.
Works where your team already works
The server is accessible in Claude, Cursor, and any MCP-compatible client. No new interface to learn, no dashboard to open. Ready to answer the question.
Touchstream's MCP server is a concrete example of what it means to make enterprise observability AI-native. The data was already there. The workflows were already defined. What Alpic added was a secure, focused interface that puts those capabilities inside the tools NOC engineers actually work in, without asking them to change how they operate.
For teams running live events, every minute of investigation time counts. The MCP server makes the judgement of your best engineer available on every shift: faster triage, more consistent incident response, and root cause surfaced across the full stack, removing friction between data and the person who needs it.
Giving engineers AI-native access to live stream data sounds straightforward until you factor in multi-tenancy and auth. Alpic understood our requirements from day one and helped us build the app that our customers needed.
Andy Webb - Product Lead at Touchstream