ReportPortal MCP server: 18 tools over test launches

What a test-reporting platform exposes to an agent, the three tools that change state, and the analytics setting that is on by default.

Last verified 1 September 2026 · from The ReportPortal repository, its release binaries and the server's own tools/list response

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the ReportPortal MCP server?

It is ReportPortal's server for the Model Context Protocol. ReportPortal aggregates automated test results, so a connected agent can read failures without opening the dashboard.

The data model is launches and test items. A launch is one execution of a suite, a test item is a case or a suite inside it, and most tools take filters over one or the other.

Three tools start ReportPortal's own analysis rather than reading data: run_auto_analysis, run_unique_error_analysis and run_quality_gate. Those are jobs on the server, not queries.

ReportPortal gives an agent the history of test launches and failures. That is evidence about test execution, not about the running application. Its CI link ends at those launches, with no agent-telemetry or cost coverage.

For an agent, ReportPortal answers what the tests did. It can find the failing item, read its log and see how it was classified. It knows nothing about the code that failed or the service in production, so it pairs with a repository connector for a fix.

What can an AI agent do with ReportPortal?

get_launches and get_last_launch_by_name ship with the same one-line description. Pin the launch by ID once found, so a later call cannot silently move to a different run.

What connecting costs before the first question

A context window is the amount of text a model can hold at once, and connecting uses some of it before the agent asks anything. Every tool the server registers sends its name, description and input schema to the model when the session starts. The tool list is charged whether or not a single tool is called.

18 tools cost between 4,146 and 7,999 tokens, about 2.1% to 4.0% of a 200,000-token window.

The schemas dominate. Descriptions average 113 characters and schemas average 885, because the filter tools pass ReportPortal's own query parameters through to the model.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted 24 August 2026 from tools/list on the released 1.3.3 macOS binary. It was run locally with a placeholder host and token.

Registration is static, so no ReportPortal instance answered. Figures cover name, description and input schema. The two models in column one share the o200k_base encoding, so they get one count. The spread is formatting, about 51%. Reasoning tokens are output, spent per request. This count is input, charged once at connection.

The Claude figures are measured rather than encoded. Anthropic publishes no local tokenizer. The definitions were sent to the API's count_tokens endpoint, and a no-tools baseline was subtracted. The range covers two tokenizer generations rather than two serializations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a third.

ReportPortal MCP server limits: where answers come back incomplete

Two limits below are about telling one launch from another; the third is about a setting that is easy to miss.

How to configure ReportPortal MCP for agents

1. 1

   Switch analytics off if that matters to you

   The binary's --help lists --analytics-off and RP_MCP_ANALYTICS_OFF. Neither appears in the README, so confirm against the version you install rather than trusting this page.

2. 2

   Decide who may delete a launch

   launch_delete and launch_force_finish register by default, and the API token is the only boundary.

3. 3

   Pick the mode deliberately

   In stdio mode RP_API_TOKEN is required; in HTTP mode it is ignored entirely and every request must carry its own bearer token.

4. 4

   Set the project

   --project avoids an agent querying a project the token can see but you did not mean.

Paste this into your agent’s instructions

Two launch tools carry the same description, so an agent can answer about a different run than the one you meant. The block makes it name the launch.

When you answer using ReportPortal MCP tools, state:
- The launch ID the answer came from, not just the suite name.
- The filters you passed. Test-item queries take ReportPortal
  filter parameters and a wrong one returns a real, wrong set.
- Whether you called launch_delete, launch_force_finish,
  update_launch or update_defect_type_for_test_items.
Never state that data does not exist. State that no items
matched the filters above.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Often you do not. ReportPortal has a documented REST API, and most of these tools pass its filter parameters through unchanged. A dashboard or a script is better served by calling it directly.

The connector earns its place during a failing build, when the agent is already in the editor and the question is which test broke and why.

At 4,146 tokens the cost of keeping it connected is small, so the argument against it is rarely context.

Connect it while triaging a failure. Call the REST API for anything scheduled or repeated.

ReportPortal MCP server setup

Download the release binary for your platform from 1.3.3, or use the Docker image the README recommends. Stdio mode needs a host, a project and an API token.

HTTP mode is different. RP_API_TOKEN is ignored, and each request must carry Authorization: Bearer <token> instead.

claude mcp add-json "reportportal" '{"command":
  "reportportal-mcp-server",
  "args":["--rp-host","https://rp.internal",
          "--project","my-project","--analytics-off"],
  "env":{"RP_API_TOKEN":"<your-token>"}}'

{"mcpServers":{"reportportal":{
  "command":"reportportal-mcp-server",
  "args":["--rp-host","https://rp.internal",
          "--project","my-project","--analytics-off"],
  "env":{"RP_API_TOKEN":"<your-token>"}}}}

[mcp_servers.reportportal]
command = "reportportal-mcp-server"
args = ["--rp-host", "https://rp.internal",
        "--project", "my-project", "--analytics-off"]
env = { RP_API_TOKEN = "<your-token>" }

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"reportportal":{"command":"reportportal-mcp-server",
  "args":["--rp-host","https://rp.internal","--project","my-project","--analytics-off"],
  "env":{"RP_API_TOKEN":"<your-token>"}}}}

{"mcpServers":{"reportportal":{"command":"reportportal-mcp-server",
  "args":["--rp-host","https://rp.internal","--project","my-project","--analytics-off"],
  "env":{"RP_API_TOKEN":"<your-token>"}}}}

{"servers":{"reportportal":{"command":"reportportal-mcp-server",
  "args":["--rp-host","https://rp.internal","--project","my-project","--analytics-off"],
  "env":{"RP_API_TOKEN":"<your-token>"}}}}

{"mcpServers":{"reportportal":{"command":"reportportal-mcp-server",
  "args":["--rp-host","https://rp.internal","--project","my-project","--analytics-off"],
  "env":{"RP_API_TOKEN":"<your-token>"}}}}

MCP_MODE=http reportportal-mcp-server \
  --rp-host https://rp.internal --analytics-off
# Clients send: Authorization: Bearer <token>

reportportal-mcp-server --rp-host https://rp.internal \
  --tls-ca-cert /etc/ssl/certs/my-ca.pem \
  --project my-project --analytics-off

Read-only access, permissions and security

• Two tools are destructive and arrive enabled

  launch_delete removes a launch and launch_force_finish closes one that is still running.

• Analytics is opt-out, not opt-in

  The server documents --analytics-off and RP_MCP_ANALYTICS_OFF to disable Google Analytics tracking, which means it is on until you do.

• Authentication changes with the mode

  In stdio mode RP_API_TOKEN is required; in HTTP mode it is ignored and the bearer header is the only credential.

• One tool uploads

  import_launch_from_file accepts inline file content and creates a launch from it, subject to the server's upload limit.

Troubleshooting

The server exits with a usage error

--rp-host is required, and the project flag is --project rather than --rp-project. The binary prints the correct flag names with --help.

HTTP mode returns 401 with a valid token

In HTTP mode RP_API_TOKEN and --token are ignored. The token has to travel per request in an Authorization: Bearer header.

The agent answered about the wrong run

get_launches and get_last_launch_by_name share the same one-line description. Have the agent pin a launch ID and state it.

TLS verification fails

Use --tls-ca-cert with your CA file, or --insecure to skip verification. The two are mutually exclusive and setting both is rejected.

ReportPortal MCP server: Reference

What engineers report

The repository is first-party, with 11 releases since it was created on 8 April 2025. Its own README and the tools/list capture are the evidence below.

Repository facts from the GitHub API, retrieved 24 August 2026. The README's own tool table was counted on the same date and matches the released binary's tools/list response.

Should you connect an agent to ReportPortal?

Yes. Best for stdio mode with --analytics-off, a project set, and a token scoped to reading unless deletion is wanted.

Strongest fit

Teams running ReportPortal who want an agent to triage a failing suite without leaving the editor.

Main advantage

Low context cost and a direct path from a launch to the log line that failed.

Main weakness

Two launch tools share a description, so the agent needs a rule about pinning an ID.

Operational risk

launch_delete and launch_force_finish register by default, with the API token as the only boundary.

Frequently asked questions