Your ElevenLabs Voice Agent Grew Up — Did You Keep Up?
We all do it: configure the agent once, it works, and we move on. The problem is that ElevenLabs doesn't rest on its laurels. In May 2026, the platform deployed four features that genuinely change the game for Quebec SMBs using a voice agent. Not gimmicks. Real tools that make the difference between an agent that answers and an agent that works for you.
Spoiler: if you haven't yet activated MCP support, content guardrails, conversation tags, and versioning, your agent is doing about 80% of what it could. This guide explains how to fix that in under an hour.
Why These 4 Features Specifically?
ElevenLabs ships updates continuously — their official changelog can be hard to follow if you're not a developer. We combed through the latest release notes and picked the four features with the best impact-to-effort ratio for a typical Quebec SMB. No coding required to take advantage of them.
Quick context: ElevenLabs raised $500 million in February 2026 at an $11 billion valuation. This isn't a startup winging it — it's the platform pushing the industry forward. IBM just announced a strategic partnership with them to deploy enterprise voice agents. What they're shipping today is what your customers will expect to find with you tomorrow.
Feature #1: MCP Support — Your Agent Can Now Act, Not Just Talk
This is probably the biggest addition. MCP (Model Context Protocol) is an open standard that lets your voice agent connect to external tools — your CRM, your calendar, your booking system — and take action on them during the conversation.
What does that look like for a Quebec SMB? Imagine your agent picks up an inbound call. The customer asks for an appointment Thursday at 2pm. With MCP, the agent checks your Google Calendar in real time, confirms availability, and creates the appointment — all without hanging up, without transferring, without human intervention. According to the ElevenLabs MCP documentation, the agent can also access your customer database to personalize the conversation.
Configuration happens directly in the ElevenLabs Conversational AI dashboard, in the Tools section. You select which MCP servers the agent can access, and define which sub-agents are allowed to call which tools. The new granular permission controls are critical here — you don't want your appointment booking agent to also have access to your billing system.
For configuration with TECHMA, we recommend starting with a single MCP tool (calendar or CRM), running 50 simulated calls, then expanding. Never connect all your tools at once — it makes debugging much harder if something goes wrong.
If you followed our tutorial on ElevenLabs Workflow Memory, MCP is the logical next step: your agent remembers the customer AND can act on your systems.
Feature #2: Content Guardrails — Protect Your Brand From Off-Topic Responses
Here's a scenario that happens more often than you'd think: a creative customer asks your voice agent about competitors, or tries to get it talking about topics completely outside its scope. Without guardrails, the agent improvises. Sometimes well. Often not.
ElevenLabs has now deployed user-configurable guardrails. You define rules in natural language — up to 10,000 characters per rule — and the platform filters responses in real time. When a rule is triggered, a guardrail_triggered WebSocket event fires, and the agent gracefully redirects the conversation.
For a Quebec SMB, here are three guardrails to activate first:
- Competition guardrail: The agent never compares your services to named competitors. If a customer mentions a competitor, the agent acknowledges hearing it and refocuses on what you offer.
- Sensitive information guardrail: The agent never discloses internal information (unpublished pricing, employee names, internal processes). Particularly important for clinics and healthcare professionals.
- Bill 25 compliance guardrail: The agent doesn't collect unnecessary personal information and consistently reminds customers how their data is handled. For everything related to Quebec's Bill 25, we've put together a complete compliance guide.
Guardrail configuration is in Agent Settings then Safety. Each guardrail can use a different evaluation model — ElevenLabs recommends lighter models for simple checks and heavier ones for complex rules, to keep latency low.
Feature #3: Conversation Tags — Finally an Organized View of Your Calls
If you're handling more than 100 calls per week, you know the problem: finding a specific conversation in the history is a nightmare. Conversation tags solve exactly that.
Since the May 7, 2026 API schema deployment, you can create custom tags, assign them to conversations, and filter your history by tag identifiers. Full operations are available via API or directly in the interface.
Concrete use cases for a Quebec SMB:
- hot-lead tag: When the agent detects a strong buying signal (price inquiry, immediate availability, budget mention), it automatically tags the conversation. Your sales team gets an alert and calls back the same day.
- technical-issue tag: For service businesses, quickly identify calls that required human escalation and analyze patterns.
- language tag: If you have a bilingual agent, automatically tag conversations in French and English to track your volume by language. We covered the specifics of bilingual agents for Quebec SMBs in this comparison.
The exclude_statuses filter is also useful: you can exclude in-progress or processing conversations from your reports, so you're only analyzing completed calls.
Feature #4: Agent Versioning — Test Without Breaking What Works
This one's going to save some sleepless nights. ElevenLabs introduced a Git-style versioning system for your agents — every modification creates a new version, and you can roll back to any previous version with one click.
Why does this matter? Because modifying a production agent's system prompt is a bit like changing a tire on the highway. We've done it, we know it can go badly. With versioning, you test the new version in parallel on a fraction of your traffic, measure results, and deploy only if metrics improve.
TECHMA's recommended process for Quebec SMBs:
- Create a new version with your changes (new prompt, new rules, new tone).
- Activate it on 10% of your inbound calls for 48 hours.
- Compare metrics: average call duration, resolution rate, customer satisfaction.
- If results are better, scale to 50%, then 100%.
- If results are worse, roll back to the previous version in under 30 seconds.
This approach completely eliminates the risk of unplanned degradation. For SMBs that have invested in optimizing their agent — like those following best practices on ElevenLabs Expressive Mode and backchanneling — being able to test improvements without risk is a paradigm shift.
Where to Start? The Recommended Activation Order
If you activate everything at once, you'll get lost in debugging. Here's the order we recommend at TECHMA:
Week 1: Activate content guardrails. This is the zero-risk change — it doesn't modify your agent's normal behavior, it only blocks edge cases. Configure at minimum the competition guardrail and the personal data guardrail.
Week 2: Set up conversation tags. Start with 2-3 tags maximum, automated via conditions in the agent workflow. Analyze your data for two weeks to identify useful patterns before adding more.
Week 3: Start using versioning. Create a baseline version of your current agent before any future modifications. Now that you have a safety net, you can test bolder improvements.
Week 4 and beyond: Connect your first MCP tool. Choose the highest-impact use case — for most Quebec SMBs, that's calendar-connected appointment booking. Measure booking success rate before and after.
What These Features Mean for Your SMB's Competitiveness
18 months ago, having a voice agent was a competitive advantage. Today, it's becoming the norm. Industry analysis shows SMB adoption of voice agents tripled in 2025-2026. Real differentiation now happens at the quality of implementation, not the simple fact of having one.
SMBs that activate MCP, guardrails, and versioning aren't just doing more AI. They're building a system that improves in a controlled way, protects their reputation, and genuinely integrates into their operations. That's the difference between a gadget and a reliable employee.
At TECHMA, we deploy these four features for our Quebec SMB clients as part of our integration engagements. If you want to know how to configure them for your specific sector, contact us for a consultation.