reportportal/reportportal-mcp-server

ReportPortal MCP Server
> **Documentation note:** This README reflects the current development branch. For the available up-to-date list of tools and capabilities, check the README on the main branch.
What is the ReportPortal MCP Server?
The ReportPortal MCP Server is a bridge between your ReportPortal instance and AI chat assistants (such as Claude Desktop, GitHub Copilot, Cursor). In simple terms, it lets you ask questions in plain English about your test runs and get answers directly from ReportPortal. It follows the official MCP guidelines.
Starting with version 2.x, the server also integrates with the **ReportPortal Test Case Management System (TMS)** — test cases, folders, milestones, and test plans — so you can browse and manage test design data through the same chat interface.
For example, instead of logging into the ReportPortal UI, you could ask your AI assistant "What tests failed in the last run?" or "List the 5 most recent test runs," and it will fetch that information from ReportPortal for you. With TMS support, you can also ask things like "List test cases in the Smoke folder" or "Create a test plan for the current sprint milestone." This makes it easy for QA testers and managers to manage test cases, query test results and test assets using natural language, speeding up planning, analysis and reporting.
Why Use It?
- **Quick Test Insights**: Instantly retrieve summaries of test runs, failure counts, or error details without writing code or navigating the UI.
- **TMS from Chat**: Browse, create, and update test cases, folders, milestones, and test plans in ReportPortal TMS without leaving your AI assistant.
- **Chat-Based Queries**: Use your favourite AI assistant (Claude, Cursor, etc.) to converse with ReportPortal data. It's like having a smart test-reporting helper in your chat window.
- **Integration Flexibility**: Works with any MCP-compatible AI tool. You simply point the tool at this server and it can run ReportPortal queries under the hood.
- **No Custom Scripts Needed**: Common queries (listing runs, getting failures, analysis, TMS lookups) are built-in as simple "commands" you invoke via chat.
Prerequisites
Before setting up the MCP server, you need the following information from your ReportPortal instance:
ReportPortal Host URL (RP_HOST)
The URL of your ReportPortal instance:
- Example: https://reportportal.example.com
- For local: http://localhost:8080
Project Key (RP_PROJECT) — Optional
This value is optional. When set, it defines the default project key used for all requests. The projectKey tool argument is only used as a fallback when no project is available from the context (i.e. RP_PROJECT env variable in stdio mode, or the X-Project HTTP header in HTTP mode).
The project key is the **unique project identifier** within the ReportPortal instance, do not use the project display name as project key. Find this on the ReportPortal general settings page:
API Token (RP_API_TOKEN)
You can get an API token from your ReportPortal Profile or generate a new one.
**Security Note:** Never commit tokens to version control or share them publicly.
Server Mode (MCP_MODE)
MCP_MODE is a system environment variable that controls the transport mode of the MCP server:
| Value | Description |
|-------|-------------|
| stdio | **(default)** Standard input/output – used for local AI tool integrations (Claude Desktop, Cursor, VS Code Copilot, etc.) |
| http | HTTP/SSE streaming – exposes an HTTP endpoint for remote or multi-client access |
This variable is only needed when you want to run the server in HTTP mode. For local AI tool integrations the default stdio mode is used and no extra configuration is required.
> **Important:** http mode is intended for server deployments. The MCP server must be **deployed and running** with MCP_MODE=http before any AI tool can connect to it remotely. See the For developers section for deployment instructions.
Installation
There are two ways to connect to the ReportPortal MCP Server:
1. **Locally** - via *Docker* (recommended) or using *pre-built binaries*.
2. **Connecting to a remote MCP server** (when the server is already deployed)
Each of these methods is suitable for any LLM provider.
Local installation
The configurations below use the default stdio mode (MCP_MODE=stdio), which is the correct choice for all local AI tool integrations. To run the server in HTTP mode instead, add MCP_MODE=http to the env block (see the For developers section for details).
Via Docker (recommended)