Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import { describe, expect, it } from 'vitest'
import type { McpServerInfo } from '../../types'
import { McpClient, staticTokenProvider } from '../client'
const server: McpServerInfo = {
name: 'scoped-server',
transport: { type: 'streamableHttp', url: 'http://scoped-server.test/mcp' },
enabled: true,
authRequired: true,
credentialRevision: 'credential-revision-1',
status: { deployed: true, ready: true },
}
describe('MCP credential revocation', () => {
it('wipes the retained bearer synchronously when a client is retired', async () => {
const client = new McpClient(server, staticTokenProvider('sensitive-upstream-bearer'))
// Simulate a live connection: connect() resolves the provider into
// currentAuthToken, which retire() must clear synchronously. Seed it
// directly so the assertion exercises retire()'s wipe, not the unconnected
// (already-undefined) state.
;(client as unknown as { currentAuthToken?: string }).currentAuthToken =
'sensitive-upstream-bearer'
const cleanup = client.retire()
expect((client as unknown as { currentAuthToken?: string }).currentAuthToken).toBeUndefined()
await expect(cleanup()).resolves.toBeUndefined()
})
})