# Getting Started with the ThousandEyes MCP Server The Model Context Protocol (MCP) is an open standard that enables AI assistants to securely connect to external data sources and tools. It provides a standardized interface for AI models to interact with your systems, making integrations portable across different AI platforms. The ThousandEyes MCP server connects AI tools to your ThousandEyes infrastructure, enabling natural language interactions with your network monitoring data. You can ask questions, run diagnostics, analyze performance, and automate workflows through conversation. By integrating ThousandEyes with your AI assistant via MCP, you can: * Use natural language queries instead of navigating the ThousandEyes platform UI or using API syntax. * Leverage intelligent analysis where the AI understands networking concepts and interprets results. * Chain multiple operations together in a single request. * Generate executive summaries, technical deep dives, or incident postmortems. * Get guided diagnostics and root cause analysis for faster troubleshooting. ### Prerequisites To use the ThousandEyes MCP server, your organization **must not** be opted out of generative artificial intelligence-powered capabilities ("GenAI") (see [ThousandEyes and GenAI](https://docs.thousandeyes.com/product-documentation/thousandeyes-and-genai)). In addition, you need: * The *API Access* ThousandEyes user permission. Any user with this permission can use the MCP server. For more information on user permissions, see [Role-Based Access Control](https://docs.thousandeyes.com/product-documentation/user-management/authorization/rb-access-control). * A ThousandEyes API token. For instructions on generating a new API token, see [User API Tokens](https://docs.thousandeyes.com/product-documentation/user-management/authorization/rb-access-control#user-api-tokens). * An MCP-compatible client, such as Claude, Cursor IDE, Microsoft CoPilot, AWS Kiro or Google Gemini. This getting started guide uses the ThousandEyes plugin for the Cursor IDE as the MCP-compatible client. For setup instructions for other supported clients, see the [ThousandEyes MCP Server integration guide](https://docs.thousandeyes.com/product-documentation/integration-guides/thousandeyes-mcp-server). ## Connecting to the ThousandEyes MCP Server To install the Cisco ThousandEyes Cursor plugin: 1. Open the **Agent Chat** window in Cursor and type `/add-plugin ThousandEyes`, then press **Enter**. 2. A browser window will open with a log in prompt. Log in to your ThousandEyes Account. 3. After logging in, you will be prompted to approve sharing your DCR proxy token with the Cursor IDE. Click **Approve**. By default, the ThousandEyes MCP tools will now be approved to run using your account permissions. To test the configuration, open the **Agent Chat** window and type: `What ThousandEyes MCP tools do I have access to?`, then press **Enter**. ## Usage ThousandEyes MCP server usage counts against your API rate limit, but the specific limit depends on your authentication method: * OAuth Bearer Token: Usage counts against your organization-wide rate limit (typically 240 requests per minute). This usage is shared with other integrations using standard API tokens. For more information, see [Rate Limits](https://developer.cisco.com/docs/thousandeyes/rate-limits/). * OAuth 2.0 Access Token: Each OAuth2 client has its own separate rate limit of 240 requests per minute. This usage does not affect your organization-wide rate limit. Unit Consumption for Instant Tests: {% hint style="warning" %} When asking the assistant to run an "Instant Test," the platform executes a live test that consumes units. The pricing is identical to scheduled tests, but is billed for a single round. For details, see [Configuration-Based Consumption Model](https://docs.thousandeyes.com/product-documentation/user-management/usage-and-billing/our-consumption-models/configuration-based-consumption-model). {% endhint %} ## Capabilities Once your AI assistant is connected to the ThousandEyes MCP server, you can use natural language to call various tools to manage tests, monitor alerts, and perform advanced analysis. For a complete list of available tools, see [MCP Server Functionality and Sample Prompts](https://docs.thousandeyes.com/product-documentation/integration-guides/thousandeyes-mcp-server#mcp-server-functionality-and-sample-prompts). Here are some examples of what you can ask your AI assistant: **Troubleshooting** * "Run an instant HTTP test to **URL** from **locations**." * Follow up: "Show me detailed results for instant test **test-id**, including response times, errors, and network path analysis. What failed?" * Follow up: "Show me all active failures in **account group** grouped by severity, with root cause analysis and impacted services." * "Analyze AWS Bedrock API health across all monitored regions: availability, latency, error rates, and active alerts." * "Customer complaints about slow checkout - show me all checkout/transaction tests with current performance vs baseline and identify bottlenecks." * "Analyze **metric** trends for **test/service** over **timeframe**. Show baseline, current values, and statistical anomalies indicating degradation." * "Compare **test/service** performance between **location A** and **location B**. Show latency, packet loss, and routing differences to identify optimal path." * "What changed in **test/service** performance around **timestamp**? Show me before/after metrics." * "Show me the network path from **location** to **target** with latency breakdown by hop." * "Is there packet loss or high latency affecting **service** right now? Show me which agents are impacted." * "Compare BGP routing for **prefix/service** across all monitors. Are there path changes?" **Summarization** * Executive summary: "Create an executive summary report outlining the health status of my monitored services and applications in **account group**? Show me availability, performance issues, and active incidents." * Morning standup: "Show overall health for the **org/account group**." * Weekly review: "Comprehensive monitoring report for last 7 days." **Optimization** * "Identify critical services and applications not currently monitored in **account group**, prioritized by business impact." * "Show me blind spots in my monitoring coverage for **account group**. Which critical services have no tests, redundancy, or multi-vantage-point validation?" * "Analyze response time baseline for **test name/ID** over the past **timeframe**. Correlate with triggered alerts, and recommend alert threshold optimizations for **account group**." * "Are there any tests that haven't run successfully in the last week?" ## Best Practices When writing prompts for your AI assistant, keep the following best practices in mind to get the most accurate and useful results: * **Be specific about time ranges**: Instead of "Show me alerts from the last six hours", use "Show me critical alerts from 2pm-8pm EST yesterday". * **Name what you want analyzed**: Instead of "Check test performance", use "Analyze HTTP response times for test 'Checkout Flow Production' over the past week". * **Specify the output format**: Instead of "Get my test results", use "Get test results as a table showing test name, status, and average response time". * **Chain operations logically**: Instead of "Find slow tests, then analyze them", use "Find all tests with >2s response time, run instant tests from five locations, compare with baseline, and create a performance report". ### Common Use Cases * **Incident response**: "We are seeing elevated error rates on our API. Run diagnostics: check current alerts, get path visualization, run instant tests from key locations, analyze service dependencies, and create an incident summary with recommended actions." * **Performance monitoring**: "Monitor my checkout flow: compare page load times week-over-week, identify bottlenecks, and alert me if any metric degrades by >20%." * **Root cause analysis**: "Walk me through debugging this intermittent connectivity issue step by step." * **SLA reporting**: "Calculate our uptime percentage for Q1 across all HTTP server tests." ## References and Additional Information * [Optimize AIOps With the ThousandEyes MCP Server)](https://www.thousandeyes.com/blog/optimize-aiops-with-thousandeyes-mcp-server) * [Cisco ThousandEyes AgenticOps: When AI Monitors AI via MCP](https://www.thousandeyes.com/blog/agentic-ops-when-ai-monitors-ai-via-mcp) * [Beyond the Chatbot: Assuring AI-powered Customer Support With ThousandEyes](https://www.thousandeyes.com/blog/ai-support-assurance) * [Monitoring AI Agents for Production Reliability](https://www.thousandeyes.com/blog/monitoring-ai-agents-production-reliability) --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.thousandeyes.com/product-documentation/getting-started/getting-started-with-the-thousandeyes-mcp-server.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.