vynix-mcp-server
Official# Vynix MCP Server
[](https://www.npmjs.com/package/@usevynix/mcp-server)
[](https://www.npmjs.com/package/@usevynix/mcp-server)
[](LICENSE)
[](https://github.com/UseVynix/vynix-mcp/actions)
[](https://glama.ai/mcp/servers/UseVynix/vynix-mcp)
Model Context Protocol server for Vynix. It gives coding agents direct access to visual feedback, bug reports, screenshots, diagnostics, comments, and issue workflows so agents can reason from real context instead of guessing.
## Why Vynix
- Feedback with evidence: page metadata, target element, screenshot, console/network context.
- End-to-end execution: inspect feedback, diagnose, generate coding prompts, create GitHub issues, update status, comment.
- Agent-safe hints: read-only/idempotent/open-world annotations for better approval behavior in MCP clients.
- Registry-ready metadata for modern MCP directories.
## Architecture
```mermaid
flowchart LR
A[MCP Client\nClaude/Cursor/Copilot/VS Code] -->|stdio or streamable-http| B[Vynix MCP Server]
B --> C[Vynix API]
B --> D[GitHub API via Vynix backend]
C --> E[Projects]
C --> F[Annotations]
C --> G[Screenshots and diagnostics]
```
## Features
- 17 production tools for read and write workflows.
- Resource catalog for server metadata, tool/prompt/skill references, and contextual summaries.
- Workflow prompts for QA, release readiness, PM briefings, and engineering planning.
- Dual transport support: `stdio` and Streamable HTTP.
- Auth via API token or email/password refresh flow.
## Installation
Node.js 18+ is required.
### NPX (recommended)
```json
{
"mcpServers": {
"vynix": {
"command": "npx",
"args": ["-y", "@usevynix/mcp-server"],
"env": {
"VYNIX_API_URL": "https://www.vynix.in",
"VYNIX_API_TOKEN": "PASTE_YOUR_TOKEN_HERE"
}
}
}
}
```
### npm global install
```bash
npm install -g @usevynix/mcp-server
vynix-mcp
```
### Docker
```bash
docker run --rm -i \
-e VYNIX_API_URL=https://www.vynix.in \
-e VYNIX_API_TOKEN=PASTE_YOUR_TOKEN_HERE \
ghcr.io/usevynix/vynix-mcp:latest
```
### Local development
```bash
git clone https://github.com/UseVynix/vynix-mcp.git
cd vynix-mcp
npm install
npm run build
npm run check
node dist/index.js
```
## Client Configuration
### Claude Desktop
Use `claude_desktop_config.json`:
```json
{
"mcpServers": {
"vynix": {
"command": "npx",
"args": ["-y", "@usevynix/mcp-server"],
"env": {
"VYNIX_API_TOKEN": "PASTE_YOUR_TOKEN_HERE"
}
}
}
}
```
Diagnostics are written to stderr; stdout is reserved for the protocol stream.
### Cursor
Use `~/.cursor/mcp.json`.
```json
{
"mcpServers": {
"vynix": {
"command": "npx",
"args": ["-y", "@usevynix/mcp-server"],
"env": {
"VYNIX_API_TOKEN": "PASTE_YOUR_TOKEN_HERE"
}
}
}
}
```
### VS Code
Use `.vscode/mcp.json` with top-level `servers`:
```json
{
"servers": {
"vynix": {
"command": "npx",
"args": ["-y", "@usevynix/mcp-server"],
"env": {
"VYNIX_API_TOKEN": "PASTE_YOUR_TOKEN_HERE"
}
}
}
}
```
### Windsurf
Use your Windsurf MCP config file with this server block:
```json
{
"mcpServers": {
"vynix": {
"command": "npx",
"args": ["-y", "@usevynix/mcp-server"],
"env": {
"VYNIX_API_TOKEN": "PASTE_YOUR_TOKEN_HERE"
}
}
}
}
```
### ChatGPT connectors
For hosted mode, use the Streamable HTTP endpoint:
- Base URL: `https://mcp.vynix.in/mcp`
- OAuth discovery: `/.well-known/oauth-authorization-server`
### Generic mcp.json
See [examples/configs/mcp.json](examples/configs/mcp.json).
## Authentication
Environment variables:
- `VYNIX_API_URL` (optional, default: `https://www.vynix.in`)
- `VYNIX_API_TOKEN` (recommended)
- `VYNIX_API_EMAIL` and `VYNIX_API_PASSWORD` (fallback login mode)
- `VYNIX_MCP_MODE` (`stdio` or `http`)
- `VYNIX_MCP_HOST`, `VYNIX_MCP_PORT`, `VYNIX_MCP_PATH` (HTTP mode)
Generate a token from: <https://www.vynix.in/mcp>
## Tools, Prompts, and Resources
- Tool reference: [docs/tools.md](docs/tools.md)
- Prompt reference: [docs/prompts.md](docs/prompts.md)
- Resource reference: [docs/resources.md](docs/resources.md)
- Skill/workflow reference: [docs/skills.md](docs/skills.md)
- Deployment guide: [docs/deployment.md](docs/deployment.md)
- Agent discovery guide: [docs/agent-discovery.md](docs/agent-discovery.md)
- LLM index file: [llms.txt](llms.txt)
- Directory listing tracker: [docs/listings-status.md](docs/listings-status.md)
## Examples
- Conversation workflows: [examples/workflows](examples/workflows)
- Prompt library (100+ prompts): [examples/prompts.md](examples/prompts.md)
## Troubleshooting
- `Not configured` error: set `VYNIX_API_TOKEN` or both `VYNIX_API_EMAIL` and `VYNIX_API_PASSWORD`.
- `401` errors: regenerate token and verify API URL.
- No tools listed: confirm the MCP config key (`mcpServers` vs `servers`) for your client.
- Hosted mode not reachable: verify `VYNIX_MCP_MODE=http` and check `/health`.
## FAQ
### Does this send data to third-party AI providers?
Only `diagnose_annotation` can invoke external AI providers through your Vynix workspace configuration.
### Is this read-only?
No. It includes read tools and write tools. MCP annotations identify mutating/open-world calls so clients can request confirmation.
### Can I self-host?
Yes. Run in stdio mode locally or HTTP mode behind your own infrastructure.
## Contributing
See [CONTRIBUTING.md](CONTRIBUTING.md). For local validation run:
```bash
npm run check
```
## Security
- Never commit API tokens.
- Prefer short-lived tokens where possible.
- See [SECURITY.md](SECURITY.md) (create one if your org requires a disclosure policy).
## License
[MIT](LICENSE)
## Changelog
[CHANGELOG.md](CHANGELOG.md)
TDQS
Scored across 17 tools
Most tools are clearly distinct, but get_metrics and get_activity both surface recent activity, which could lead to misselection. Other tools like list_annotations vs get_annotation or get_annotation_analysis vs diagnose_annotation are clearly differentiated by read/action semantics.
All tool names follow a consistent verb_noun snake_case pattern (e.g., list_projects, update_annotation_status, diagnose_annotation). There is no mixing of conventions or vague verbs, making the set predictable and easy to navigate.
With 17 tools, the server is slightly above the ideal range, but each tool serves a distinct purpose in the annotation management, AI diagnosis, and GitHub integration workflow. The count feels justified rather than padded, though a few tools could potentially be consolidated.
The core lifecycle is well covered: listing and getting annotations, updating status, commenting, running/reading AI analysis, generating prompts, and creating GitHub issues. Missing operations like create/delete annotation or update/delete comments are minor gaps, likely because annotations are created externally or those actions are outside the server's scope.