Drive OctoPerf from Your AI Assistant
The OctoPerf MCP Server connects Claude, Codex, Gemini, Copilot and any AI agent straight to your load-testing platform.
Import virtual users, fix replay errors, run scenarios and read back metrics - all in plain language, without leaving the chat.
Validation: 3 actions failing with 401 - the
sessionToken is replayed as a static value.sessionToken from the login response.Validation is now green: 24/24 actions pass.
Run completed: 0.2% errors, median response time down 12% vs last night - the new cache pays off.
→ Full bench report
Your AI Agent Now Speaks Load Testing
The OctoPerf MCP Server is a thin, stateless bridge between an AI agent and the OctoPerf REST API, built on the open Model Context Protocol. Anything you can do from the OctoPerf web UI, your agent can do on your behalf - once you authorize it.
No glue code, no scripts to maintain: point your assistant at https://api.octoperf.com/mcp, log in once, and start working. Every tool call is authenticated as you and resolves to a regular OctoPerf API call, so there's nothing new to secure.
One Connection, the Whole Workflow
The server exposes around 100 tools covering the full performance-testing lifecycle.
Every tool returns a deep-link to the matching OctoPerf UI page, so your agent hands you a clickable URL whenever it summarizes a result.
How It Works
From a plain-language request to actionable results, in five steps:
You ask
"Auto-correlate this virtual user and re-run the validation."
The agent picks
the right OctoPerf tools for the job.
The server executes
them against the OctoPerf API - authenticated as you.
OctoPerf responds
with live data, metrics and deep-links.
The agent explains
the result in plain language and suggests the next step.
Works With the Tools You Already Use
Because it speaks the open Model Context Protocol over Streamable HTTP, the server plugs into any MCP-aware client:
- Claude.ai - add it as a Custom Connector
- Claude Code - one-command install via the official OctoPerf plugin
- Claude Desktop - add the URL to your config
- Codex -
codex mcp add, then log in with OAuth - Gemini CLI -
gemini mcp add, then/mcp auth - GitHub Copilot (VS Code) - agent mode via
.vscode/mcp.json
One server, every assistant. As the MCP ecosystem grows, your integration grows with it.
Secure by Design
OAuth 2.1 + PKCE + Dynamic Client Registration. No API keys to store or rotate; every call carries a short-lived token minted from your identity.
Every tool call runs with your own permissions - no service accounts, no shared secrets. Revoke any agent anytime from Account → Connected applications.
A Head Start in Claude Code
Install the official OctoPerf plugin and get the MCP server, the agent guide and eight ready-made workflow skills in a single step:
/plugin marketplace add OctoPerf/octoperf-claude-plugins
/plugin install octoperf@octoperfThe bundled skills turn common testing chores into one-line requests:
Just Ask
No DSL to learn, no API reference to memorize. Plain language is the interface:
"Import this HAR as a new virtual user."
"Auto-correlate the session token and re-validate."
"Run the Checkout scenario on the cloud provider."
"Is the test still running? Give me the live error rate."
"Compare the last two runs and tell me what regressed."
"Export the bench report as a PDF I can share."
Get Started in Minutes
From zero to your first AI-driven load test:
Have an OctoPerf account
The free plan is enough to try it.
Add the server
Paste the endpoint into your MCP client, or install the Claude Code plugin.
Log in once
The first call opens a browser tab for OAuth login.
Start asking
Point your agent at the agent guide and let it work.
MCP endpoint
https://api.octoperf.com/mcpClaude Code plugin
/plugin install octoperf@octoperfRunning OctoPerf Enterprise on-premise?
The MCP server self-hosts alongside your instance with a bundled OAuth IdP. Self-hosted setup →
Bring AI Into Your Performance Testing