Customize
MCP servers
Connect Model Context Protocol servers: tools, resources and prompts.
Faelith speaks the Model Context Protocol. A connected server's tools appear as mcp__<server>__<tool>, its Markdown resources as skills, and its prompts as slash commands (/mcp__github__pr_review).
Configure#
.faelith/mcp.json in a project (or mcpServers in ~/.faelith/settings.json for every project):
{
"mcpServers": {
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "${GITHUB_TOKEN}" }
},
"acme": {
"type": "http",
"url": "https://mcp.acme.example/mcp",
"headers": { "Authorization": "Bearer ${ACME_MCP_TOKEN}" }
}
}
}Sources merge in this order, later wins: user settings → project .faelith/mcp.json (or .mcp.json) → .faelith/mcp.local.json → --mcp-config <file>.
Server fields#
| Field | Meaning |
|---|---|
command, args, env | Local stdio server; command is the executable, flags go in args |
url, type, headers | Remote server: http (Streamable HTTP, default), sse or websocket |
oauth | clientId, callbackPort, scopes for OAuth servers; tokens are kept in the OS credential store |
disabled | Keep the entry but do not start it |
${VAR} expands from the environment, so tokens never need to be in the file. Remote servers must use HTTPS (plain HTTP only for localhost).
Manage#
/mcp open the MCP panel
/mcp status connection state and tool counts
/mcp list every mcp__ tool
/mcp add <name> <command> [args...]
/mcp remove <name>
/mcp reconnect <name>Trust and safety#
Servers defined by a project stay blocked until you trust that workspace (complete onboarding there). MCP tools pass through the same mode gate, permission engine and secret scan as built-in tools, and local servers run inside the session sandbox. Treat servers with side effects as write tools.
Found a mistake or a gap? Tell us and we will fix the page.