MCP servers and skills
Extend your agents with Model Context Protocol servers and Agent Skills.
MCP servers and skills extend what your agents can do. MCP servers add external tools and data sources (browsers, error trackers, analytics). Skills let agents work with other modalities — video, audio, images, and speech. Both are configured in Project Settings and apply to every agent in the project.
MCP servers
Configure Model Context Protocol (MCP) servers to extend agent capabilities with custom tools and integrations. MCP servers allow agents to interact with external services, databases, and APIs during implementations. MCP server settings apply to all agents.
Playwright MCP
Playwright MCP gives agents the ability to control a headless browser. With it enabled, agents can:
- Navigate to your app's live preview
- Click links and buttons, fill out forms
- Take screenshots of the UI and send them back to you in the chat
- Check the JavaScript console for errors
This is useful when you want agents to verify that UI changes actually work. Instead of trusting that the code looks right, the agent can test it in a real browser and see what users would see.
Screenshots taken by agents appear as Artifacts in the chat. This makes it easy to review visual changes without running the app yourself.
To enable: Go to Project Settings → Skills and MCPs → Enable Playwright MCP.
Sentry MCP
Sentry MCP lets agents query your Sentry account directly. With Sentry MCP enabled, agents can look up actual errors with full stack traces, request data, and get user context.
To enable:
- Go to Project Settings → Skills and MCPs → Enable Sentry MCP
- Add your Sentry Access Token
- Generate a token at sentry.io/settings/account/api/auth-tokens
- Recommended scopes for full Sentry MCP compatibility:
org:read,team:read,project:read,project:releases,event:read,team:write,project:write,event:write
Your token is stored encrypted and only passed to agents when Sentry MCP is enabled.
PostHog MCP
PostHog MCP lets agents query your PostHog account directly. With PostHog MCP enabled, agents can look up feature flags, insights, events, logs, and error tracking data — useful for debugging issues or understanding how a feature is being used in production.
To enable:
- Go to Project Settings → Skills and MCPs → Enable PostHog MCP
- Add your PostHog Personal API Key
- Generate a key using the MCP Server preset at app.posthog.com/settings/user-api-keys?preset=mcp_server
Your key is stored encrypted and only passed to agents when PostHog MCP is enabled.
Custom MCP servers from your repos
Superconductor supports project-scoped MCP configuration stored in your repos. If your repo contains an MCP config file, Superconductor merges it into the agent's MCP configuration automatically. This lets you version-control MCP servers alongside your code so every agent gets the same tools without manual setup.
Each agent reads from its own config file format:
| Agent | MCP config file | Other merged config |
|---|---|---|
| Claude Code | [repo]/.mcp.json | Hooks, plugins, and marketplaces from [repo]/.claude/settings.json; commands and agents from [repo]/.claude/ |
| Codex | [repo]/.codex/config.toml | — |
| Gemini CLI | [repo]/.gemini/settings.json | Commands |
| OpenCode | [repo]/opencode.json | Commands, agents, plugins |
| Amp | [repo]/.amp/settings.json | — |
When there are conflicts between your repo config and the config Superconductor generates (e.g., from MCP servers enabled in Project Settings), Superconductor's config takes precedence.
Skills
Skills are built on an open standard and work with any Superconductor agent. Enable them from Project Settings → Skills and MCPs. Once a skill is enabled, just ask your agent in the chat to generate an image, analyze a video, etc, and agents will use the relevant skill.
Available skills
- Video Understanding - Analyze videos up to an hour long. Supports local files (mp4, mov, webm, avi, mkv) and YouTube links.
- Audio Understanding - Transcribe and analyze audio up to 9.5 hours long. Supports mp3, wav, m4a, ogg, flac, and YouTube links.
- Image Generation - Generate images from text descriptions. Can use existing images or videos as style references.
- Video Generation - Create 4-8 second videos from text prompts. Supports image-to-video animation with native audio.
- Text to Speech - Convert text to natural-sounding speech. 30 voices, 24 languages, automatic language detection.
All skills require a Google Gemini API key configured at the workspace or project level.
Custom skills from your repos
You can also add custom skills by including them in your repo. Superconductor picks up agent-specific skill directories and makes them available:
| Agent | Skills directories |
|---|---|
| Claude Code | [repo]/.claude/skills/ or [repo]/.agents/skills/ |
| Codex | [repo]/.codex/skills/ or [repo]/.agents/skills/ |
| Gemini CLI | [repo]/.gemini/skills/ or [repo]/.agents/skills/ |
| OpenCode | [repo]/.opencode/skills/ or [repo]/.agents/skills/ |
| Amp | [repo]/.agents/skills/ |
The [repo]/.agents/skills/ path is part of the Agent Skills open standard and works across all agents. If multiple repos define a skill with the same name, the first repo (alphabetically) wins.
Skills use cases
Some examples of what you can do with Agent Skills:
- Create visual database structure diagrams automatically
- Analyze competitor demo videos to produce competitive analysis reports
- Generate multiple landing page redesign mockups from text descriptions
- Identify bugs from customer call recordings
- Extract customer testimonials from YouTube videos for your website