dvera-mcp

DVERA — CT Verification Skills for C/C++ Coding Agents

한국어: README.ko.md

DVERA (Dynamic Verification Agent) brings CT C/C++ verification work into coding agents such as Claude Code, Codex CLI, Cursor, and GitHub Copilot. It uses the working context of source code, requirements, build information, source changes, and existing test assets to prepare and request CT verification work.

CT (Controller Tester) is a test automation solution for unit, integration, and code-based testing of mission-critical C/C++ software. It brings together test-environment setup, test design and generation, execution, code-coverage analysis, reporting, and traceability management.

DVERA does not replace CT's verification results. The coding agent helps prepare and coordinate verification work; CT performs the analysis, build, test execution, and result aggregation. Generated tests must be reviewed for test intent, data, and expected results, then confirmed with actual execution results and measured coverage.

What the CT and DVERA workflow adds

Verification need	CT and DVERA workflow
Complex C/C++ environments	Capture compiler, macro, include-path, toolchain, build, and target conditions before analysis.
Unit and integration verification	Design and execute tests at function, module, and interface levels, including existing GoogleTest assets where applicable.
Reviewable evidence	Connect test intent, execution results, statement/branch/MC/DC coverage, reports, and requirement links for review.
Repeated change verification	Re-run affected tests after code changes and review new failures and coverage deltas for regression follow-up.
AI-assisted test work	Use code, requirements, and change context to prepare test work, then improve it with CT analysis and measured results.

Workflow at a glance

1. Prepare the compiler, macro, include-path, toolchain, and target context for a CT project.
2. Analyze the project and resolve environment or configuration issues through a controlled feedback loop.
3. Generate, execute, and improve unit or integration tests using build, execution, and coverage feedback.
4. Review coverage and results, produce reports, and re-run affected verification after source changes.

The Skills in this repository document that product-connected workflow. They are not standalone replacements for a CT installation.

Install

In Claude Code:

/plugin marketplace add SuresoftTechnologies/dvera-plugin
/plugin install dvera@suresofttech

Or from a terminal:

claude plugin marketplace add SuresoftTechnologies/dvera-plugin
claude plugin install dvera@suresofttech

Skills then load as /dvera:ct-init-project, /dvera:ct-test-loop, and so on. The plugin adds about 620 tokens to every session; each Skill body is read only when that Skill runs.

Product-backed actions need a deployed CT environment. Without one, a Skill stops at its installation check and shows how to reach product support.

Example requests

Trigger examples from the CT user guide. Responses can vary with the AI model and conversation context; if the result differs from what you expect, rephrase the request with more detail and retry.

Request	Skill
Create a CT project and start verification.	ct-init-project
Run tests in CT.	ct-init-project
Generate tests for the calculate_checksum function.	ct-test-loop
Find functions without tests and generate tests for them.	ct-test-loop
Run regression tests.	ct-regression
Open CT.	ct-open

Skills

All Skills require a deployed CT environment for their product-backed actions and stop with an installation notice when none is found.

Skill	CT workflow stage
ct-init-project	Start or prepare a CT verification project.
ct-extract-macro	Capture target compiler macros and build-environment facts.
ct-make-conf	Prepare the analysis configuration for the selected toolchain.
ct-setup-project	Create the CT project and apply build inputs.
ct-analysis-loop	Analyze the project and resolve configuration or toolchain issues.
ct-test-loop	Generate, run, and improve tests.
ct-kb-update	Record verified workflow observations after a completed CT run.
ct-regression	Re-run existing tests after a code change and review deltas.
ct-self-healing	Run regression evidence collection and identify CT Self-Healing review candidates.
ct-run-gtest	Re-run GoogleTest assets registered in CT and review results and coverage.
ct-req-to-test	Generate and execute a CT AI test from approved requirements.
ct-report	Export execution and coverage evidence.
ct-open	Open the CT project for visual inspection.
ct-orchestrator	Resume the appropriate stage of an in-progress verification workflow.

Product-backed workflow

DVERA works with a deployed CT and DVERA environment. This repository documents the verification workflow and ships an MCP server that forwards tool calls to a local CT installation, but does not include the CT runtime or product installation.

Without the deployed environment, users can inspect this repository's documented workflow, but a Skill that needs CT actions stops at its installation check. It must not claim that CT analysis, execution, coverage measurement, or reporting has occurred.

MCP tools

Give the agent CT's verification tools directly. With the MCP server (scripts/dvera-mcp.py) connected, it can create the project, run analysis, generate and execute tests, and read coverage back without leaving the conversation - the same tools CT exposes to its own agent. Calls go to CT's ct_tool.py entry point, and the server keeps no process of its own between them.

Connect it in one of two ways.

Point an MCP client at the interpreter CT already ships - no extra install:

{
  "mcpServers": {
    "dvera": {
      "command": "C:/Program Files/Suresoft/CT 2026/python/python.exe",
      "args": ["/path/to/dvera-plugin/scripts/dvera-mcp.py"]
    }
  }
}

On Linux the interpreter is at <CT install>/python/python3. Any Python 3.8 or later works if you would rather use your own. Replace the script path with wherever this repository sits: clone it, or use the copy the plugin manager installed (claude plugin list shows where plugins live). The server takes the source directory from the client's working directory, so run the client from the project you want to verify.

Or install the bundle from this repository's releases: download dvera-mcp.mcpb and open it with a client that installs MCP bundles. It resolves its own Python, so nothing has to be installed first.

Set CT_HOME if CT is installed outside the default location. The server is safe to leave configured on any machine: it starts either way and simply offers no tools until CT is there.

Requirements for product-backed verification

• CT 2026.06 or later
• A valid CT license
• A supported C/C++ build and target environment
• DVERA integration supplied as part of the deployed CT environment

For a product demo, purchase, or deployment consultation, contact bizcenter@suresofttech.com.

Privacy Policy

The MCP server in this repository runs entirely on your machine.

• It sends nothing to Suresoft Technologies or to any third party. There is no telemetry, no analytics, and no network call of its own.
• Tool arguments are passed to the CT installation on the same machine, and CT's response is returned to your MCP client. Nothing is written outside CT's own workspace.
• It stores no credentials and reads no files beyond the CT installation directory and the paths a tool call names.
• Anything CT itself records - workspaces, analysis output, reports - is governed by CT's own handling of that data, not by this repository.

The Skills are documentation: they carry no code that runs on your machine.

The full policy is in PRIVACY.md. Questions about data handling in the CT product itself go to bizcenter@suresofttech.com.

License and trademarks

This repository is licensed under the MIT License. The license covers everything in this repository - the documentation, the Skill files, and the MCP server source. It does not cover CT itself: CT and DVERA are Suresoft Technologies products, and no product functionality or product licence is granted by this repository.

GoogleTest is a trademark of Google LLC. This repository is not affiliated with, sponsored by, or endorsed by Google. Other product names may be trademarks of their respective owners.