Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Description from the maker (original English)
Polyglot MCP test generation with 8-section contract enforcement across Python, TS, Java, Go, C++.
These three are never merged into one score: opinions, measurements and popularity are different things.
We have not written a plain-language summary for this one yet. Below are the facts we read from the registry.
Pick the AI app you use, then copy the settings.
You need
Which app do you use?
Pick your app above. We remember it, so next time the button says “Install in …”.
Open Claude Desktop → Settings → Developer → Edit Config, and paste this into claude_desktop_config.json
{
"mcpServers": {
"universal-test-framework": {
"command": "uvx",
"args": [
"universal-test-framework",
"--transport",
"stdio"
]
}
}
}Run this in your terminal
claude mcp add universal-test-framework -- uvx universal-test-framework --transport stdioPaste into .cursor/mcp.json in your project, or ~/.cursor/mcp.json
{
"mcpServers": {
"universal-test-framework": {
"command": "uvx",
"args": [
"universal-test-framework",
"--transport",
"stdio"
]
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
Paste into .vscode/mcp.json in your project
{
"servers": {
"universal-test-framework": {
"type": "stdio",
"command": "uvx",
"args": [
"universal-test-framework",
"--transport",
"stdio"
]
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
This one runs on your computer, and ChatGPT can only connect to servers on the internet. Pick another app.
Your choice stays on this device in a cookie; no account needed.
These settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.
Runs on your computer; the tool list appears in your AI app after you install it.
We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 7, 2026
The registry name was verified with the GitHub account phoenice-labs, which owns the repo
We have not read this repo from GitHub yet
We have not read this repo from GitHub yet
This one runs on your computer; there is no internet endpoint
Runs on your computer and uses local settings instead
Runs on your computer; we can only read tool lists from internet endpoints
Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.
It passed 1 of 1 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.
The registry does not list a required key, though some servers ask you to sign in on first use.
It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).
The GitHub repo last changed –, and the latest version (1.0.2) reached the registry 7 months ago.
Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need Python (uv) on your computer.
In the same category: Spec Driven Development, Remote Desktop Commander, FactLot
Servers in the same category from other makers, sorted by stars.
Reading a card:★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Cited CRE facts and deterministic calculators for agents: rent, CAM, debt sizing, IRR, parcel IDs.
Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.
Contract-driven test enforcement and reporting for LLM-generated code — via MCP, VS Code, Copilot CLI, Claude, Cursor, or Python SDK.
UTF is an MCP server and contract enforcement engine. It does not replace your AI coding assistant — it gives every test your AI writes a mandatory quality gate, a structured audit trail, and a comprehensive compliance report.
AI tools — GitHub Copilot, Claude, Cursor, Gemini — generate tests fast. The problem: fast ≠ trustworthy.
| Without UTF | With UTF |
|---|---|
| Tests exist but nobody knows *why* | Every test is linked to a requirement |
| "Covers everything" — nobody can prove it | Traceability matrix maps REQ → test |
| CI passes; real behavior untested | Gap analysis flags what is NOT covered |
LLM wrote a test that asserts True | Meaningfulness check rejects trivial assertions |
| No history of what was tested | Persistent registry survives session restarts |
| Report shows pass/fail counts only | 8-section per-test detail cards in HTML report |
You may already use markdown files (AGENTS.md, copilot-instructions.md, .cursorrules) to guide your AI. UTF is complementary — not competing:
| Markdown instructions | UTF |
|---|---|
| Describe *how* to write tests | Enforce a contract on *every* test produced |
| Rely on LLM to follow instructions | Block tests that fail the contract at generation time |
| No verification after generation | Score each test 0–1 against 8 measurable criteria |
| No persistent state between sessions | SQLite registry persists all tests and results |
| No compliance report | HTML report with per-test 8-section detail cards |
Use markdown instructions to *shape* how your AI thinks. Use UTF to *verify* and *report on* what it produced.
TC-{PRJ}-{MODULE}-{NNN} and classic TC-{MODULE}-{NNN} both accepteduvx (zero-clone), local clone, Docker, GitHub MCP Registry, pip SDK.utf/rules/ override global defaults| # | Method | How Started | Registry Location | @utf Slash Cmds | Best For |
|---|---|---|---|---|---|
| 1 | uvx (zero-install) | mcp.json + uvx | /.utf/utf.db | ❌ use natural language | Any developer with uv |
| 2 | Local Clone | install-vscode-mcp.ps1 | /.utf/utf.db | ✅ with extension | Contributing / customizing UTF |
| 3 | Docker / HTTP | docker compose up | Named volume or bind mount | ❌ use natural language | Team / remote / CI |
| 4 | GitHub MCP Registry | Auto via client | /.utf/utf.db | ❌ use natural language | Discoverable via MCP marketplace |
| 5 | pip + SDK | Python import | Caller controls cwd | N/A | CI scripts / programmatic |
All methods (1, 2, 4) using stdio transport write the registry to `${workspaceFolder}/.utf/utf.db` — isolated per project and persistent across sessions.
Requires uv. No cloning, no venv.
Add to %APPDATA%\Code\User\mcp.json (Windows) or ~/.config/Code/User/mcp.json (macOS/Linux):
{
"servers": { "utf": { "type": "stdio", "command": "uvx", "args": ["--from", "universal-test-framework", "utf-server", "--transport", "stdio"], "cwd": "${workspaceFolder}", "env": { "UTF_PROJECT_DIR": "${workspaceFolder}" } } } }
Registry:
.utf/utf.dbinside${workspaceFolder}— isolated per project, persistent across sessions.
Reload VS Code after editing
mcp.json.
git clone https://github.com/phoenice-labs/Universal-Test-Framework
cd Universal-Test-Framework
.\scripts\install-vscode-mcp.ps1
.\scripts\install-vscode-mcp.ps1 -InitProject -ProjectDir C:\my-project
Reload VS Code → open Copilot Chat → ask naturally: generate e2e tests for my backend.
The install script writes this entry to mcp.json:
{
"utf": { "type": "stdio", "command": " ", "args": ["-m", "mcp_server.server", "--transport", "stdio"], "cwd": "${workspaceFolder}", "env": { "PYTHONPATH": " " } } }
Registry:
${workspaceFolder}/.utf/utf.db— isolated per project.
`@utf` slash commands are available after the VSIX extension is installed.
# Start the UTF server
cd Universal-Test-Framework docker compose -f docker/docker-compose.yml up -d
Connect from VS Code by adding to mcp.json:
{
"servers": { "utf-remote": { "type": "sse", "url": "http://localhost:8765/sse" } } }
Registry: persisted in a named Docker volume (
utf_registry → /app/.utf/utf.db).
For per-project isolation with Docker, use a bind-mount in
docker-compose.yml:
`yaml
volumes:
- /path/to/your/project/.utf:/app/.utf
`
Because Docker uses HTTP/SSE transport (no
${workspaceFolder}templating), passproject_dir
explicitly in tool calls, or set
UTF_PROJECT_DIRin the container environment.
Per-project Docker workflow:
# docker-compose.override.yml
services: utf-server: volumes:
Once published, UTF is discoverable via the MCP marketplace. Clients that support server.json install it automatically. The generated mcp.json entry is equivalent to Method 1 (uvx).
Registry isolation: The MCP registry schema does not support a
cwdfield at the registry level.
UTF resolves project isolation via (in priority order):
1.
project_dirargument passed to each tool call
2.
UTF_PROJECT_DIRenvironment variable
3.
Path.cwd()fallback (server's working directory)
For correct isolation, ensure the MCP client writes
"cwd": "${workspaceFolder}"and
"UTF_PROJECT_DIR": "${workspaceFolder}"in the generated entry (UTF'smcp-gallery.jsondoes this).
pip install universal-test-framework
from mcp_server.tools.generate_tests import generate_tests
from pathlib import Path
result = generate_tests( test_type="unit", source_code=open("src/auth.py").read(), project_dir=str(Path.cwd()), # ← pass explicitly for correct registry isolation ) print(result["suite_code"])
Registry:
/.utf/utf.dbwhenproject_diris passed; falls back toPath.cwd().
| Tool | Description |
|---|---|
generate_tests | Generate a complete test suite satisfying the 8-section contract |
validate_test_contract | Validate any test (generated or hand-written) against the contract |
analyze_coverage | Identify coverage gaps in an existing test suite |
build_traceability_matrix | Build a requirements → tests traceability matrix |
suggest_test_types | Recommend test types with rationale from source code |
detect_language_framework | Auto-detect programming language and test framework |
import_test_results | Import JUnit XML from any test run into the UTF registry |
query_registry | Query the persistent per-project test registry |
run_tests | Execute test files and return structured CI results |
run_mutation_tests | Run mutation testing and return mutation score |
feedback_status | Get gap analysis, coverage health, and trend report |
generate_report | Generate HTML / JUnit / JSON contract compliance report |
health | Server health check: version, uptime, tool count |
Every test generated by UTF must satisfy all 8 sections. Tests that fail any hard section are blocked — returned in blocked_tests, never silently included.
| # | Section | What It Must Contain | Weight |
|---|---|---|---|
| 1 | test_id | Unique ID: TC-{TYPE}-{NNN} (e.g. TC-US-042) | 10% |
| 2 | why_generated | Rationale tied to a requirement (≥50 chars) | 10% |
| 3 | requirement_mapping | At least one US-, AC-, REQ-, JIRA-, BUG-, or NFR- reference | 15% |
| 4 | how_it_exercises | GIVEN / WHEN / THEN with inputs, mocks, assertions (≥100 chars) | 20% |
| 5 | coverage_contribution | Coverage type + module + estimated % | 15% |
| 6 | expected_outcome | Precise return values, status codes, state changes (≥50 chars) | 15% |
| 7 | gaps_missing | Honest list of what this test does NOT cover (≥40 chars) | 10% |
| 8 | meaningfulness_check | Self-assessment: meaningful / redundant / hallucinated (≥50 chars) | 5% |
Minimum passing score: 0.85. Tests below this threshold are blocked regardless of individual section presence. "none" in gaps or vague rationale like "to test the function" are rejected.
| Language | Frameworks | Test Types |
|---|---|---|
| Python | pytest | unit, integration, api, e2e, security, performance |
| TypeScript | Jest, Vitest, Playwright | unit, integration, e2e, api |
| JavaScript | Jest, Vitest | unit, integration |
| Java | JUnit 5 + AssertJ, Maven | unit, integration, api |
| Go | go-test + testify | unit, integration, api |
| C++ | Google Test | unit |
This is the canonical flow for using UTF with any AI CLI (Copilot, Claude, Cursor) or VS Code. Follow the phases in order — skipping Phase 1 registration means the report has no Per-Test Contract Detail cards.
┌─────────────────────────────────────────────────────────────────────────┐
│ PHASE 1 — CONTRACT GENERATION (LLM writes, UTF validates + registers) │ │ │ │ ① generate_tests (scaffold) │ │ ② LLM writes real test methods — each with its own TC-ID and │ │ 8-section comment block (WHY / REQ / HOW / COV / OUT / GAP / MEAN)│ │ ③ validate_test_contract — score must be ≥ 0.85 per test │ │ ④ register_contracts — parse test file, upsert status=generated rows │ │ ⑤ generate_report — verify Per-Test Contract Detail cards appear │ │ │ │ PHASE 2 — EXECUTION (pytest/jest/maven runs, results captured) │ │ │ │ ⑥ pytest --junit-xml=utf-tests/reports/results.xml │ │ ⑦ import_test_results — upsert executed/failed rows
Only the beginning is shown; read the rest in the repo