Introducing the Testmo MCP: your test repository, connected to your AI assistant

This release, we’re publicly kicking off a closed beta for the Testmo MCP. The Testmo MCP connects your QA data to the AI agents, harnesses, and AI-enabled IDEs your team already uses to make test context available for AI coding agents. Meanwhile, it also makes Testmo accessible by your AI assistants to help you accelerate and automate your existing QA processes in Testmo.

What the Testmo MCP is

The Testmo MCP is a hosted MCP server built on the Model Context Protocol (MCP) standard. No download or local hosting required: simply add the Testmo MCP to any AI assistant that supports MCP like Claude Code, Claude Desktop, Cursor, VS Code, Codex, Windsurf, etc to start working with your Testmo data right from your AI-enabled workspace. Once set up, your AI can read from — and, when you allow it, write to — Testmo in plain language.

In practice that means your assistant can work with your projects, milestones, runs, results, sessions, and case repository without you leaving your editor or copy-pasting between windows. Your tests stop being a separate silo and start acting as a live knowledge base for both QA and coding.

Introducing Tests as Context in the AI SDLC

One of the primary problems Testmo MCP addresses is a familiar one for anyone involved in software development in 2026.

AI coding assistants are good at producing plausible code. They’re much less good at knowing the rules your team learned the hard way: the validation limits, the status transitions, the awkward edge cases that live in your test cases and, often, nowhere else. Ask an assistant to change how discount codes work at checkout and it will happily write something that looks right and quietly break three behaviours you already test for.

The Testmo MCP puts that context in front of the assistant before the code is written. It allows your AI coding agents to reference acceptance criteria refined in your test cases as well as regressions you caught in past release cycles of having to rely on limited context from Jira stories and code repositories alone.

Finding the right tests: Intelligent Search

One of the most powerful capabilities included in this update of the Testmo MCP is the addition of the search_cases tool.

When you ask your AI assistant about your repository, the search_cases tool allows it to use Testmo’s Intelligent Search engine to search a vectorized database of your entire Testmo test repository with a combination of semantic and keyword-based search.

This gives your AI agents the ability to:

Reference your entire test repository at once, without having to reference individual test cases one by one
Search across meaning and keyword, allowing your AI agent to find the most relevant test cases based on your prompt
Significantly improve the context for downstream workflows without burning through as many AI credits

A quick reminder: the same way a precise query in any search box beats a vague one, the more specific your prompt, the more relevant test case context you will be able to surface via the MCP.

Tools to accelerate your entire QA workflow

In addition to the search_cases tool, Testmo’s MCP comes with 64 other tools that span your Testmo projects, case repository, milestones, manual runs & results, exploratory sessions, and automated runs.

These tools leverage Testmo’s APIs to enable your AI assistant to reference your Testmo test data, and if you choose ‘write’ privileges during MCP setup, even automate a number of the steps for your existing QA workflow like:

Generating new test cases
Updating existing test cases
Creating new milestones
Creating new runs
Uploading results
Referencing cases, milestone, runs, and results for custom reporting
And more!

Specialized workflows with skills in the Testmo Claude plugin

Beyond tools, this update to the MCP supports the use of a new set of skills available via the Testmo Claude plugin. (Coming soon to other AI assistants!)

These skills instruct your AI assistant about how to use the Testmo MCP for more powerful, specialized workflows, like:

Spec Implementer — before you write code, it pulls the cases that describe how the relevant area is supposed to behave and hands them to the assistant as guardrails, so the code respects rules your tests already encode.
Regression Preventer — describe a change you’re about to make and get a short brief of the existing cases it could break, before you’ve written a line.
Change Evaluator — after a change, get a per-case prediction of pass, fail, or uncertain, so you know where to focus manual testing before you push.
Case Importer — create cases in Testmo from a spreadsheet or an export from another tool. (This one writes to your repository, so it needs read-write access and Python available locally.)

A caveat that applies across all of them: these are assistants, not oracles. Change Evaluator predicts outcomes by reasoning over your cases — it doesn’t execute your tests. Treat the output as a fast first read and confirm with a real run before you ship.

You control the access scope

When you set up your MCP connection, you can decide how much the MCP can do by setting the access scope accordingly:

Read-only limits your AI assistant to search and read your repository — cases, folders, runs, results, sessions — but change nothing. It’s the right default for analysis, coverage review, and using tests as coding context.
Read-write also lets the assistant create and update Testmo data, which workflows like Case Importer need.

Connection uses OAuth. Your inputs and any AI-generated outputs are treated as Customer Content: never shared with other customers, and not used to train underlying models. The MCP reads only your Testmo repository — your source code or any other data you give your assistant never leaves your own AI environment unless you deliberately share it in a prompt.

What this means for teams

For teams using AI coding assistants day to day: your repository becomes context the assistant draws on before it writes code, rather than documentation it never sees. The rules your tests encode travel into the work automatically, so you catch the regression at the design stage instead of after the pull request.

For QA teams with a substantial repository: Test Selection makes years of accumulated cases searchable by meaning, not just exact keywords — so the coverage you already have is easier to find, reuse, and reason about when you’re deciding what to test.

For teams evaluating AI carefully: you start read-only, connect over OAuth, and your content is never used to train models. You can try it without handing over write access or a line of source code.

Getting access to Testmo MCP

While the Testmo Claude Plugin is now available via the Claude Plugin Marketplace, the Testmo MCP itself is available to selected teams while we tune it toward general availability.

If you’d like to Testmo MCP access, please fill out the Testmo MCP Beta signup form →

As we roll the MCP out to more cohorts Testmo, we’ll reach you with more information about how to set up the MCP with connection details and the beta skills. If you’re already using AI coding assistants and have a substantial test repository to draw on, you’re exactly who we want to hear from.

Also in this release

While MCP is the headline, this release included a few other additions and improvements for all Testmo users too.

New Team and Business plan sizes

We’ve added two new plans so it’s easier to pick a size that fits your team:

Team (5 users) — a smaller entry point for compact teams.
Business (15 users) — a mid-tier step before larger plans.

Session management over the API

In July we rounded out the API for milestones and results. This month sessions get the same treatment, so you can create, update, and delete exploratory testing sessions programmatically:

Create — POST /projects/{project_id}/sessions
Update — PATCH /sessions/{session_id}
Delete — DELETE /sessions/{session_id}

That completes full CRUD for sessions over the API, so session tracking can be wired into your own automation the same way runs, milestones, and results already are.

Other API additions

Retrieve a single folder — a new GET /folders/{folder_id} returns one repository folder by ID.
Project-scoped run updates — a new PATCH /projects/{project_id}/runs/{run_id} updates a run within a project scope, and runs now accept case_ids to control which cases are included, on both create and update.
Consistent custom field responses — single-select (dropdown) and multi-select custom fields now return {id, name} in session and run result responses, so the two types return the same shape.
Clearer content field handling — send an unprefixed content key (say description instead of custom_description) and the API now returns a clear 422 with a hint to use the custom_ prefix, rather than silently discarding the value.

As always, thanks for your feedback — a lot of what’s here came directly from it. Full details are in the changelog, and the API reference has been updated for every new endpoint. If you’ve got a view on where the MCP should go next, sign up for the beta and let us know about it! Testmo MCP Beta