CredVault

Pragma MCP Servers

MCP (Model Context Protocol) lets Pragma connect the AI agent to external tools and data sources in a controlled way.

Users manage MCP from Settings → MCP Servers (and related AI settings).

What MCP Means For You

An MCP server is a tool bridge. Once connected, Comp / the Pragma agent can call that tool during a task — for example to read a service, query an API, or use a specialized workflow helper.

MCP is powerful. Treat each server like an integration with permissions.

How To Add Or Enable MCP

  1. Open Settings → MCP Servers.
  2. Browse available servers or add a configured server.
  3. Enable only the servers you need for the current project.
  4. Restart or reconnect if the UI asks you to after changing server config.

Some setups support file-based MCP servers (config living with your project or machine). Prefer project-local config when the tools are project-specific.

What You Should Know Before Enabling A Server

  • Scope: Only enable tools you trust with the current workspace.
  • Secrets: Do not hard-code API keys in prompts. Use Pragma Secrets or environment variables.
  • Autonomy: Higher agent autonomy + powerful MCP tools can make larger changes. Start with ask-first mode until you trust the setup.
  • Audit: If something unexpected happens, disable the server and review recent agent actions.

File-Based MCP Servers

File-based MCP configuration is useful when:

  • A team shares the same tool wiring through repo config
  • You want different tools per project
  • You need reproducible agent setups across machines

Keep MCP config out of public forks if it contains private endpoints or credentials.