Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
/**
* Integration: channel-reader → mcp-host
*
* Validates the RPC forward path from channel-reader to mcp-host.
* NOTE: mcp-host POST /v1/runtime/messages is synchronous — it waits for the
* full LLM response before returning. Tests use a short AbortSignal (8s) so
* they verify the endpoint ACCEPTS the connection without waiting for LLM
* completion. Full LLM flow is covered by the bash E2E suite.
*
* Requires: minikube cluster with mcp-host port-forwarded on :8080
* make minikube-pf-all
*/
import { beforeAll, describe, expect, it } from 'vitest'
import { issueRealRpcToken } from '../mcp-host/runtimeAuth.js'
import {
EXTERNAL_REST_API_URL,
MCP_HOST_URL,
RPC_PROXY_URL,
fetchJson,
isServiceUp,
} from './helpers.integration.js'
let mchUp = false
let runtimeAuthReady = false
let runtimeToken = ''
beforeAll(async () => {
mchUp = await isServiceUp(MCP_HOST_URL)
if (!mchUp) {
console.log('[channel-reader-mcp-host] mcp-host not available — tests will be skipped')
return
}
const externalUp = await isServiceUp(EXTERNAL_REST_API_URL)
const rpcProxyUp = await isServiceUp(RPC_PROXY_URL)
if (!externalUp || !rpcProxyUp) {
console.log(
'[channel-reader-mcp-host] external-rest-api/rpc-proxy not available — auth-required tests will be skipped'
)
return
}
runtimeToken = await issueRealRpcToken(
['host:message:invoke', 'host:status:read'],
[process.env.TEST_HOST_REF ?? 'chatllm']
)
runtimeAuthReady = true
})
function makeMessage(overrides: Record<string, unknown> = {}) {
return {
content: 'integration test ping',
hostRef: process.env.TEST_HOST_REF ?? 'chatllm',
channelId: 'integration-test',
sender: 'integration-test-user',
channelType: 'telegram',
timestamp: new Date().toISOString(),
messageId: `int-${Date.now()}`,
...overrides,
}
}
/**
* Posts to mcp-host with a short timeout (8s).
* Returns the HTTP status, or 202 if the request times out (meaning
* the endpoint accepted the connection but LLM is still processing).
*/
async function postMessageQuick(
body: unknown,
token = runtimeToken
): Promise<{ status: number; timedOut: boolean }> {
try {
const res = await fetch(`${MCP_HOST_URL}/v1/runtime/messages`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
Authorization: `Bearer ${token}`,
},
body: JSON.stringify(body),
signal: AbortSignal.timeout(8000),
})
return { status: res.status, timedOut: false }
} catch (err: unknown) {
if (err instanceof Error && (err.name === 'AbortError' || err.name === 'TimeoutError')) {
// Request was accepted by the server — LLM just took longer than 8s.
// This is expected behavior for mcp-host; count as "accepted" (202-like).
return { status: 202, timedOut: true }
}
throw err
}
}
// ── Health ────────────────────────────────────────────────────────────────────
describe('channel-reader → mcp-host: health', () => {
it('/v1/runtime/health returns 200', async () => {
if (!mchUp) return
const { status } = await fetchJson(`${MCP_HOST_URL}/v1/runtime/health`)
expect(status).toBe(200)
})
})
// ── Message endpoint acceptance ───────────────────────────────────────────────
describe('channel-reader → mcp-host: POST /v1/runtime/messages', () => {
it('rejects a valid message without runtime auth', async () => {
if (!mchUp) return
const res = await fetch(`${MCP_HOST_URL}/v1/runtime/messages`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify(makeMessage()),
signal: AbortSignal.timeout(5000),
})
expect(res.status).toBe(401)
})
it('accepts a valid message (2xx — LLM may still be processing)', async () => {
if (!mchUp || !runtimeAuthReady) return
const { status } = await postMessageQuick(makeMessage())
// 200 = LLM completed within 8s; 202 = timed out (LLM still processing)
// Both are acceptable — the endpoint is reachable and accepted the message
expect([200, 202]).toContain(status)
})
it('returns 400 immediately for a message with invalid JSON body', async () => {
if (!mchUp || !runtimeAuthReady) return
const res = await fetch(`${MCP_HOST_URL}/v1/runtime/messages`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
Authorization: `Bearer ${runtimeToken}`,
},
body: '{ invalid json',
signal: AbortSignal.timeout(5000),
})
expect(res.status).toBe(400)
})
it('does not return 5xx for a message missing optional fields', async () => {
if (!mchUp || !runtimeAuthReady) return
const { status } = await postMessageQuick({
content: 'minimal message',
hostRef: 'chatllm',
channelId: 'test',
sender: 'test-user',
channelType: 'telegram',
timestamp: new Date().toISOString(),
messageId: `int-minimal-${Date.now()}`,
})
expect(status).toBeLessThan(500)
})
})
// ── Status endpoint ───────────────────────────────────────────────────────────
describe('channel-reader → mcp-host: GET /v1/runtime/status', () => {
it('rejects status without runtime auth', async () => {
if (!mchUp) return
const { status } = await fetchJson<Record<string, unknown>>(`${MCP_HOST_URL}/v1/runtime/status`)
expect(status).toBe(401)
})
it('returns 200 with a non-empty status object', async () => {
if (!mchUp || !runtimeAuthReady) return
const { status, data } = await fetchJson<Record<string, unknown>>(
`${MCP_HOST_URL}/v1/runtime/status`,
{
headers: {
Authorization: `Bearer ${runtimeToken}`,
},
}
)
expect(status).toBe(200)
expect(Object.keys(data).length).toBeGreaterThan(0)
})
})