TestingOps.ai

Connection guides · Connection guide

Claude and TestingOps: your test runs, cases and defects in a Claude chat

Connect Claude to the TestingOps MCP server, then ask why last night's run failed, turn acceptance criteria into stored test cases, start a run or file a defect from the conversation, with the permissions of the person who connected it.

  • Connection guide
  • Setup guide
Request a demo Anthropic: custom connectors with remote MCP
Connection
Connection guide
Writes back
Depends on the access you grant
Last checked
7 Nov 2026

In short

Claude and TestingOps in short

Claude can use TestingOps through the TestingOps MCP server, whose endpoint ends in /api/mcp/v1, and then calls TestingOps tools for runs, cases, defects and the knowledge base. In claude.ai and Claude Desktop the endpoint is added as a custom connector that signs in on a TestingOps consent page, while Claude Code can also send a TestingOps access key in the Authorization header.

Claude acts as the person who connected it, in one company, with read only or read and write access; the Claude setup is a guide, and TestingOps installs no connector in Claude.

Because claude.ai and Claude Desktop call custom connectors from Anthropic's cloud, a self-hosted TestingOps must be reachable from the internet over https for them, while Claude Code connects from the developer's own machine.

What it does

What it does for your testing

  1. 1

    A failed run explained in the chat

    Input
    A question such as "which cases failed last night, and why?"
    TestingOps
    Lists recent runs, returns the failed results of the one you mean and analyses the failures without writing anything
    Output
    Failures grouped by cause, with likely product defects separated from environment or test problems
  2. 2

    Acceptance criteria become stored cases

    Input
    A story's acceptance criteria pasted into the chat
    TestingOps
    Checks the inventory for similar cases first, then creates the scenario and its cases in the project
    Output
    Draft cases with steps and expected results in TestingOps, waiting for a person to approve them
  3. 3

    A run started and followed

    Input
    The cases to run and the test environment
    TestingOps
    Starts the run in the background and returns its status and captured evidence whenever Claude asks
    Output
    Results with the screenshots, video, logs and traces the run captured
  4. 4

    A fix checked against later runs

    Input
    A defect marked as fixed
    TestingOps
    Checks whether runs that happened after the fix pass for the cases linked to it
    Output
    A confirmed fix, or a clear statement that no run since the fix confirms it

Step by step

How it works

  1. Copy the endpointUnder Integrations → API & MCP, the MCP server card shows the Endpoint (streamable HTTP). Every assistant uses this same address.
  2. Add it to ClaudeIn claude.ai or Claude Desktop, add a custom connector with the address and keep sign-in on. In Claude Code, add an HTTP MCP server with an access key header, or leave the header out and sign in with /mcp.
  3. Approve access in TestingOpsThe consent page asks which company or customer Claude may use and whether it may read only or also write, and shows the address the browser returns to. Do not approve an address you do not recognise.
  4. Claude calls TestingOps toolsWhen a conversation needs it, Claude calls tools for runs, cases, defects, data sets and the knowledge base. Every call runs with your permissions, in that one company.
  5. Review or end accessSign-ins appear under Connected apps and keys under Access keys. Removing the app or revoking the key ends access at once.

The connection runs one way: Claude calls TestingOps. Outside tools that TestingOps agents call themselves are a separate card, Extra tools your agents may use (outgoing, over MCP). Claude's own menus are documented by Anthropic and can change.

Connect

Setup

  1. In TestingOps, open Integrations → API & MCP and copy the Endpoint (streamable HTTP) from the MCP server card. The Claude card under Connection guides opens the same steps in the product's Help.
  2. claude.ai or Claude Desktop: Customize → Connectors → Add → Add custom connector. Give it a name, paste the endpoint, keep sign-in on, then turn it on in a chat from the + menu. On Team and Enterprise plans the owner adds it once in the organisation's connector settings and every member signs in.
  3. Claude Code: run claude mcp add --transport http testingops <endpoint> --header "Authorization: Bearer <key>" with a key from Access keys → Create key, or leave out the header and sign in with /mcp.
  4. On the TestingOps consent page, pick the company or customer and read only or read and write, then approve.
  5. To end access, remove the app under Connected apps or revoke the key under Access keys.

Documentation: Anthropic: custom connectors with remote MCP · Claude Code: connect tools through MCP

Where you connect it in TestingOps: Integrations → API & MCP → Connection guides → Claude

Use cases

Use cases

Morning triage of a nightly run

Trigger
A QA lead opens Claude after the nightly regression run
Uses
Run list, failed results and the failure analysis
Agent activity
Claude reads the latest run, groups the failures by cause and separates environment problems from likely defects
Result
A triage list to act on without opening each result
Condition
Read only access is enough

From acceptance criteria to cases in the project

Trigger
A tester pastes a story's acceptance criteria into Claude
Uses
Duplicate check, scenario and case creation
Agent activity
Claude checks the inventory for existing cases, then writes the scenario and its cases into the chosen project
Result
Draft cases in TestingOps that a person reviews and approves
Condition
Needs read and write access

A developer verifies a fix from the terminal

Trigger
A developer marks a defect fixed while working in Claude Code
Uses
Run start and status, defect fix verification
Agent activity
Claude Code starts the cases linked to the defect, follows the run and checks the fix against runs after it
Result
A fix confirmed by a later run, or a note that it is not confirmed yet
Condition
A run against a production environment needs a person's approval

Capabilities

Capabilities and limits

What is read

  • Test cases, scenarios and data sets in the projects the connecting person can open
  • Runs with results, captured evidence and comparisons between runs
  • Defects, traceability, release readiness and the project knowledge base
  • Issues and wiki pages in trackers connected to TestingOps, through that connection

What is produced

  • Claude's answers, built from TestingOps tool results
  • Failure analysis, release readiness and change impact results when Claude asks for them

Where it can write

Only with read and write access: scenarios and cases (created as drafts), runs (start or cancel), defects and their status, data sets, and issues or wiki pages in connected trackers. A read only grant cannot change anything.

How it connects

A custom connector in claude.ai or Claude Desktop, or an HTTP MCP server in Claude Code, pointing at the TestingOps MCP endpoint (streamable HTTP, JSON-RPC)

Account and permissions

  • Sign-in on the TestingOps consent page, choosing the company or customer and read only or read and write; the grant is listed under Connected apps
  • Or, for Claude Code, an access key sent as Authorization: Bearer <key>, Read only or Read and write, expiring after 30 days to 1 year
  • Claude acts as the connecting person; a customer account reaches only that customer's projects

Limits

  • Tools only: the TestingOps MCP server offers no resources or prompts
  • claude.ai and Claude Desktop need a TestingOps address reachable from Anthropic's cloud over https
  • Runs against a production environment need the production run permission and a person's approval
  • How many custom connectors a Claude plan allows is set by Anthropic

Example

Example: how a team uses it

Illustrative scenario, not a customer story.

A team runs its web regression suite every night. In the morning the QA lead asks Claude which cases failed and why. Claude reads the run and its failure analysis: most failures share one cause, a test environment that timed out during login, and two point to a changed validation message on the checkout page.

The lead asks Claude to file a defect for the message change. With read and write access, Claude files it in TestingOps linked to the failing result, with the expected and actual text, and leaves the timeouts for a rerun once the environment is back.

Frequently asked questions

Is this the same as running TestingOps agents on Claude models?

No. This guide lets Claude call TestingOps as a tool. Choosing Anthropic as the model provider that TestingOps' own agents run on is a separate setting under Integrations → AI models.

Which projects can Claude see?

The ones the connecting person can open, in the one company chosen on the consent page. A customer account reaches only that customer's projects; a grant from an internal team account covers every customer of the company, and the consent page says so before you approve.

Can Claude change our data?

Only with read and write access. Cases it creates start as drafts that a person approves, and a run against a production environment needs a person's approval. With read only access TestingOps refuses every write.

Can our Claude team share one connector?

On Claude Team and Enterprise plans the owner adds the connector once and each member signs in with their own TestingOps account, so everyone keeps their own permissions. Anthropic sets how many custom connectors each plan allows; the Free plan allows one.

Does it work with a self-hosted TestingOps?

claude.ai and Claude Desktop call custom connectors from Anthropic's cloud, so the TestingOps address must be reachable from the internet over https; Anthropic documents which of its addresses to allow for servers on private networks. Claude Code connects from the developer's machine and can use an instance only that machine reaches.

How do we end Claude's access?

Remove Claude under Connected apps, or revoke its key under Access keys; access ends immediately. Each person sees their own apps and keys there, and company admins see everyone's.

Next step

Request a demo to see Claude answer questions about your own runs and cases through TestingOps.

Request a demoMore connection guides integrations