Skip to main content
MCP servers and Skills expand what your Agents can do and know. Both are managed in the Config Center, and both are enabled per Agent.

​
MCP servers

MCP (Model Context Protocol) servers give Agents extra tools. Vibex supports three transports:

​
Install from the market

Open Config Center → MCP; the MCP market opens by default:
  1. Search or page through the grid. A card shows the transport, source, command summary, and enabled Agents.
  2. Click Install, then review the command, transport, environment, and headers that will be saved.
  3. Choose which Agents receive the server.
  4. After installation the entry appears in the installed list; you can Reinstall or edit it.
Market search queries the upstream catalog as you type. Review the source and command before installing; a market entry is not an executable endorsed by Vibex.

​
Add manually

  1. Click New MCP server in the MCP list.
  2. Fill in Name and choose a transport.
  3. Fill in Command / Arguments / Environment, or URL / Headers.
  4. Choose a scope: User or Workspace.
  5. Save, then click Validate on the resource card.
  6. Under Agent enablement, turn on the Agents that should use it.
You can also use Import existing MCP to read candidates from another Agent’s native configuration. Import shows a preview and never silently overwrites.
Validation is offline: it checks required fields and URL format (HTTP/SSE must start with http:// or https://), but starts no process and makes no network request. Passing validation does not prove the server is reachable.
MCP tools may read files, access the network, or change external systems. Confirm the server’s source and permissions before enabling it for an Agent.

​
Skills

A Skill is a SKILL.md plus any referenced assets, giving an Agent reusable domain knowledge or a workflow. Sources include:

​
Install from the market

Open Config Center → Skills; the Skill market opens by default:
  1. Search or page through the grid. A card shows the source, install count, official marker, and enabled Agents.
  2. Click Install, then read the full SKILL.md body, file list, and refused files before committing.
  3. Choose which Agents receive the Skill.
  4. After installation you can Reinstall; removed entries explain themselves first.

​
Add manually

  1. Open Config Center → Skills and click New Skill or Import existing Skills.
  2. Fill in Name, Description, and the Skill body (the SKILL.md text).
  3. Save and Validate.
  4. Under Agent enablement, turn on the Agents that need it.
Once configured, type $ in the composer to find and insert an enabled Skill.
A Skill’s body and files enter the Agent’s context. Read what you install, especially from the market; do not install one that asks for credentials or runs unknown commands.

​
Prompts

Prompts are another reusable resource in the Config Center, for instructions you type repeatedly:
  • Create one under Config Center → Prompts with a name and body.
  • Type / in the composer to search and insert it.
  • A Prompt can be enabled per Agent.
  • A Prompt is plain text and carries no files or tools; use a Skill when you need those.

​
Export to Agent-native configuration

MCP, Skills, and Prompts can be written into an Agent’s own configuration files so external CLIs see the same setup. On a resource card, use Preview export → Apply → Rollback:
  1. Preview export lists the files that will change or be created.
  2. Review them, then Apply.
  3. If something goes wrong, Rollback from the same place.
The target depends on the Agent, for example Codex’s ~/.codex/config.toml, Claude’s ~/.claude/settings.json, and each Agent’s Skills directory.
Always review the preview before you Apply. By default Vibex does not rewrite an Agent’s native home configuration; it writes only when you explicitly apply.

​
Scope and security

Use the smallest scope so project-specific tools do not leak into every project. MCP servers currently offer User and Workspace scopes.
  • Disable before deleting; disabling keeps the record so you can restore it.
  • Deleting a resource does not delete files already written into an Agent’s native directory; use the export preview’s rollback.
  • Logs and diagnostics must omit tokens, private keys, environment values, prompts, and file contents.
  • Market installs write into the runtime home; do not enable an untrusted market entry for every Agent.