AI Agents Overview¶
!!! info "TL;DR" Canviq's MCP server lets AI agents programmatically create surveys, query responses, analyze sentiment, and trigger workflows. Built for Claude Desktop and other MCP-compatible AI systems.
What is the MCP Server?¶
The Model Context Protocol (MCP) server is a plain HTTP + JSON endpoint that exposes Canviq's survey and feedback capabilities as tools AI agents can call. This enables agentic workflows like:
- Automated survey creation — AI creates targeted surveys based on product milestones or user behavior
- Real-time sentiment analysis — Claude Haiku analyzes open-text responses for sentiment trends
- Workflow orchestration — Trigger automated actions when survey responses meet conditions
- Feedback triage — Automatically categorize, tag, and route submissions to the right team
Use Cases¶
Product Team Automation¶
An AI agent monitors your product analytics. When it detects a drop in retention, it:
- Creates a 3-question survey targeting churned users
- Distributes via email to the last 100 churned accounts
- Analyzes responses for common themes
- Creates GitHub issues for the top 3 pain points
- Posts a summary to Slack
All without human intervention.
Support Deflection¶
When a user submits feedback about a known issue, the AI agent:
- Detects similarity to existing submissions using embeddings
- Links the new submission to the existing one
- Auto-responds with a status update and ETA
- Adds the user to a notification list for that issue
Roadmap Planning¶
Before quarterly planning, the AI agent:
- Queries all submissions with 20+ votes
- Runs sentiment analysis on comments
- Groups by category and theme
- Generates a prioritized list with justification
- Creates draft roadmap cards in your project management tool
How It Works¶
The MCP server exposes a tool registry. Each tool is a discrete operation like survey_create or get_pmf_score. AI agents authenticate with API keys and call a tool by sending a plain JSON body to POST /api/mcp/tools:
{
"tool": "survey_create",
"arguments": {
"title": "User Onboarding Experience",
"questions": [
{ "type": "rating", "text": "How likely are you to recommend us?" }
]
}
}
The server validates scopes, applies rate limits, executes the tool, logs the action, and returns the result. Canviq does not parse free text itself. An MCP-compatible client (Claude Desktop, Claude Code, Cursor) is responsible for turning a user's plain-English request into the structured call above before it reaches this endpoint.
Agents authenticate with a key generated from the Agent Management page (/admin/settings/agents), not the SDK page or the generic API Keys wizard. See Authentication → Key types for why those other key types don't work here.
Key Features¶
- Scope-based permissions — Fine-grained access control per agent
- Audit logging — Every action logged for compliance (SOC 2, ISO 27001)
- Rate limiting — Distributed rate limits via Upstash Redis
- Durable execution — Workflow engine powered by Trigger.dev
Planned, not yet implemented: describing a request in plain English and real-time (SSE) update streams are both on the roadmap but do not exist today. See MCP Tools API → Planned Features before building an integration that depends on either.
What's Next¶
- MCP Server Architecture — HTTP + JSON transport, tool registry pattern
- Authentication — API keys, scopes, policies
- Available Tools — Complete reference of MCP tools
- Workflows — Automated actions and triggers