feat(mcp): Tasks extension (io.modelcontextprotocol/tasks) (#1283)

* feat(mcp): speak spec revision 2026-07-28 (stateless core)

Adopt the 2026-07-28 MCP spec revision on the connector endpoint while
keeping every handshake-era client (2025-06-18 and earlier) byte-identical:

- Accept per-request _meta protocol negotiation
  (io.modelcontextprotocol/protocolVersion); unsupported versions return
  UnsupportedProtocolVersionError (-32022) with the supported list.
- Implement server/discover (spec MUST): supported revisions, capabilities
  including the extensions field, identity, instructions, freshness hints.
- Decorate results for stateless clients: required resultType, serverInfo
  in _meta, and CacheableResult ttlMs/cacheScope on tools/list,
  prompts/list, resources/list, resources/read.
- Validate the standard Mcp-Method/Mcp-Name request headers when present
  (HeaderMismatchError -32020); absence stays accepted.
- Declare the ratified MCP Apps extension (io.modelcontextprotocol/ui) in
  capabilities; the widgets already use the ratified mime type and
  _meta.ui.resourceUri shape, so no widget changes are needed.
- OAuth: include the RFC 9207 iss parameter on every authorization
  response (success and error) and advertise
  authorization_response_iss_parameter_supported in RFC 8414 metadata.

Resource-not-found already used -32602 and tools/list ordering was already
deterministic; both are covered by the new test file.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(mcp): Tasks extension (io.modelcontextprotocol/tasks)

Durable handles for long-running MCP tool calls, per the official Tasks
extension. A client that declares the extension in its per-request
capabilities gets a CreateTaskResult (resultType: "task") immediately;
the work completes after the response via after() and lands in the new
mcp_tasks table for tasks/get polling. Clients that did not declare the
extension are never handed a task (spec MUST).

- New mcp_tasks table (migration 20260729094000): company-scoped SELECT
  RLS, service-role-only writes (mirrors pending_operations), 1-hour
  expiry, status lifecycle CHECK. pg-real coverage included; triaged as
  excluded in the full-archive backup contract (transient state).
- tasks/get (creator-scoped), tasks/cancel (cooperative, working-only
  flip), tasks/update (ack no-op: no input_required flows yet).
- Tool opt-in via shouldRunAsTask predicate; first producer is
  gnubok_audit_package, the one genuinely long-running blocking call
  (multi-minute ZIP generation). estimate_only stays synchronous.
- Tool failures complete the task with the standard isError envelope,
  exactly what the synchronous call would have returned; the failed
  status stays reserved for infrastructure errors.
- server/discover and initialize now advertise the tasks extension.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(mcp): creator-only task RLS, enforced expiry sweep, RoPA entry

Compliance-swarm follow-ups on the mcp_tasks migration (editing the
migration is safe: it has not shipped beyond the ephemeral PR preview):

- SELECT RLS tightened from company-wide to auth.uid() = user_id so the
  DB grant matches the creator-scoped tasks/get contract; task results
  carry raw tool output (Art. 5(1)(c)). pg test now proves a same-company
  colleague cannot read the row.
- The 1-hour retention is now enforced, not aspirational: createMcpTask
  opportunistically deletes expired rows on every creation
  (idx_mcp_tasks_expires), best-effort (Art. 5(1)(e)).
- RoPA entry mcp.async_task_handles added to .compliance/ropa.yaml
  (Art. 30).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(mcp): literal terminal-update payload for the phantom-column guard

The conditional spreads in resolveMcpTask made the payload unresolvable
for the no-phantom-columns guard (362 > 360 ceiling). A literal payload
writing null for absent terminal fields is equivalent here: the terminal
transition sets the complete terminal state.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Jakob Wennberg
2026-07-29 18:15:06 +02:00
committed by GitHub
parent 4501f118c2
commit ef25a87d75
8 changed files with 785 additions and 5 deletions
+37
View File
@@ -466,6 +466,43 @@ processing_activities:
- writes_staged_for_explicit_user_approval
- plaintext_personnummer_blocked_from_staging
- id: mcp.async_task_handles
name: MCP-uppgiftshandtag för långkörande verktygsanrop
purpose: >-
Låta en behörig MCP-klient hämta resultatet av ett långkörande
verktygsanrop (t.ex. generering av arkivpaket) via ett hållbart
uppgifts-id i stället för att hålla anslutningen öppen. Raden lagrar
verktygets svar tills klienten hämtat det; endast den skapande
användaren kan läsa, och utgångna rader rensas löpande vid nya anrop.
lawful_basis: art_6_1_b
special_category_basis: null
controller: gnubok-tenant
processor: supabase
data_subjects:
- business_owner
- company_member
data_categories:
- user.financial
- user.activity_timestamp
recipients:
- name: Supabase
country: EU
role: processor
international_transfers:
applicable: false
mechanism: null
note: EU-only processor.
retention:
duration: 1h
basis: transient_tool_call_state_storage_limitation
stored_in:
- mcp_tasks
security_measures:
- rls_user_scoped_select # mcp_tasks SELECT: auth.uid() = user_id
- service_role_only_writes
- creator_scoped_task_polling # tasks/get requires the creating user
- expired_rows_purged_on_creation
- id: arsredovisning.signature_evidence
name: Underskriftsbevis för årsredovisning
purpose: >-
@@ -0,0 +1,300 @@
/**
* Tests for the MCP Tasks extension (io.modelcontextprotocol/tasks):
* capability gating, CreateTaskResult, post-response completion writes,
* and the tasks/get / tasks/update / tasks/cancel methods. Does NOT re-test
* audit-package generation itself (covered by audit-package.test.ts).
*/
import { describe, it, expect, vi, beforeEach } from 'vitest'
import { eventBus } from '@/lib/events/bus'
import { isTaskCapableClient, TASKS_EXTENSION_ID } from '../tasks'
const mocks = vi.hoisted(() => ({
scopes: [] as string[],
taskInserts: [] as Record<string, unknown>[],
taskUpdates: [] as Record<string, unknown>[],
taskRow: null as Record<string, unknown> | null,
}))
vi.mock('@/lib/auth/api-keys', async (importOriginal) => {
const actual = await importOriginal<typeof import('@/lib/auth/api-keys')>()
mocks.scopes = [...actual.ALL_SCOPES]
const membershipChain: unknown = new Proxy(
{},
{
get(_t, prop) {
if (prop === 'then') {
return (resolve: (v: unknown) => void) =>
resolve({
data: { company_id: '11111111-1111-4111-8111-111111111111', role: 'owner' },
error: null,
})
}
return () => membershipChain
},
}
)
const makeChain = (result: unknown): unknown =>
new Proxy(
{},
{
get(_t, prop) {
if (prop === 'then') {
return (resolve: (v: unknown) => void) => resolve(result)
}
if (prop === 'single') {
return () => Promise.resolve(result)
}
return () => makeChain(result)
},
}
)
const tasksBuilder = {
delete: () => makeChain({ data: null, error: null }),
insert: (row: Record<string, unknown>) => {
mocks.taskInserts.push(row)
return {
select: () => ({
single: async () => ({
data: {
id: 'task-1',
company_id: row.company_id,
user_id: row.user_id,
tool_name: row.tool_name,
status: 'working',
status_message: null,
result: null,
error: null,
poll_interval_ms: 2000,
ttl_ms: 3600000,
created_at: '2026-07-29T09:00:00.000Z',
},
error: null,
}),
}),
}
},
update: (patch: Record<string, unknown>) => {
mocks.taskUpdates.push(patch)
return makeChain({ data: null, error: null })
},
select: () =>
makeChain(
mocks.taskRow
? { data: mocks.taskRow, error: null }
: { data: null, error: { message: 'not found' } }
),
}
return {
...actual,
extractBearerToken: vi.fn().mockReturnValue('test-token'),
validateApiKey: vi.fn().mockResolvedValue({
userId: 'user-1',
companyId: '11111111-1111-4111-8111-111111111111',
scopes: mocks.scopes,
apiKeyId: 'key-1',
apiKeyName: 'Test key',
mode: 'live',
}),
createServiceClientNoCookies: vi.fn(() => ({
from: (table: string) => {
if (table === 'mcp_tasks') return tasksBuilder
if (table === 'company_members') return membershipChain
// fiscal_periods (and everything else) resolves to no rows, so the
// audit-package execution fails fast with "Fiscal period not found".
return makeChain({ data: null, error: { message: 'not found' }, count: 0 })
},
})),
}
})
import { handleMcpRequest } from '../server'
const TASK_META = {
'io.modelcontextprotocol/protocolVersion': '2026-07-28',
'io.modelcontextprotocol/clientCapabilities': {
extensions: { [TASKS_EXTENSION_ID]: {} },
},
}
function mcpRequest(method: string, params?: Record<string, unknown>): Request {
return new Request('http://localhost:3000/api/extensions/ext/mcp-server/mcp', {
method: 'POST',
headers: { 'Content-Type': 'application/json', Authorization: 'Bearer test-token' },
body: JSON.stringify({ jsonrpc: '2.0', id: 1, method, ...(params ? { params } : {}) }),
})
}
async function readBody(request: Request): Promise<{
result?: Record<string, unknown>
error?: { code: number; message: string }
}> {
const response = await handleMcpRequest(request)
const body = await response.json()
return { result: body.result, error: body.error }
}
/** The after() fallback runs the job as a floating promise: let it settle. */
async function settleBackgroundWork(): Promise<void> {
await new Promise((resolve) => setTimeout(resolve, 0))
await new Promise((resolve) => setTimeout(resolve, 0))
}
describe('MCP Tasks extension', () => {
beforeEach(() => {
vi.clearAllMocks()
eventBus.clear()
mocks.taskInserts.length = 0
mocks.taskUpdates.length = 0
mocks.taskRow = null
})
describe('capability detection', () => {
it('detects the tasks extension in per-request capabilities', () => {
expect(isTaskCapableClient(TASK_META as Record<string, unknown>)).toBe(true)
expect(isTaskCapableClient({})).toBe(false)
expect(
isTaskCapableClient({
'io.modelcontextprotocol/clientCapabilities': { extensions: {} },
})
).toBe(false)
})
it('advertises the tasks extension in server/discover', async () => {
const { result } = await readBody(mcpRequest('server/discover'))
const capabilities = result?.capabilities as { extensions: Record<string, unknown> }
expect(capabilities.extensions[TASKS_EXTENSION_ID]).toEqual({})
})
})
describe('CreateTaskResult', () => {
it('returns a task handle to a task-capable client and completes after the response', async () => {
const { result } = await readBody(
mcpRequest('tools/call', {
name: 'gnubok_audit_package',
arguments: { fiscal_period_id: '22222222-2222-4222-8222-222222222222' },
_meta: TASK_META,
})
)
expect(result?.resultType).toBe('task')
const task = result?.task as Record<string, unknown>
expect(task.taskId).toBe('task-1')
expect(task.status).toBe('working')
expect(typeof task.pollIntervalMs).toBe('number')
expect(typeof task.ttlMs).toBe('number')
expect(mocks.taskInserts).toHaveLength(1)
await settleBackgroundWork()
// The mocked DB has no fiscal period, so the execution fails and the
// task completes with the standard isError envelope.
expect(mocks.taskUpdates).toHaveLength(1)
expect(mocks.taskUpdates[0].status).toBe('completed')
const stored = mocks.taskUpdates[0].result as Record<string, unknown>
expect(stored.isError).toBe(true)
expect(stored.resultType).toBe('complete')
})
it('never returns a task to a client that did not declare the extension', async () => {
const { result } = await readBody(
mcpRequest('tools/call', {
name: 'gnubok_audit_package',
arguments: { fiscal_period_id: '22222222-2222-4222-8222-222222222222' },
_meta: { 'io.modelcontextprotocol/protocolVersion': '2026-07-28' },
})
)
expect(result?.resultType).not.toBe('task')
expect(result?.isError).toBe(true)
expect(mocks.taskInserts).toHaveLength(0)
})
it('keeps estimate_only synchronous even for task-capable clients', async () => {
const { result } = await readBody(
mcpRequest('tools/call', {
name: 'gnubok_audit_package',
arguments: {
fiscal_period_id: '22222222-2222-4222-8222-222222222222',
estimate_only: true,
},
_meta: TASK_META,
})
)
expect(result?.resultType).not.toBe('task')
expect(mocks.taskInserts).toHaveLength(0)
})
})
describe('tasks/get', () => {
it('returns the working state while the task runs', async () => {
mocks.taskRow = {
id: 'task-1',
company_id: '11111111-1111-4111-8111-111111111111',
user_id: 'user-1',
tool_name: 'gnubok_audit_package',
status: 'working',
status_message: null,
result: null,
error: null,
poll_interval_ms: 2000,
ttl_ms: 3600000,
created_at: '2026-07-29T09:00:00.000Z',
}
const { result } = await readBody(mcpRequest('tasks/get', { taskId: 'task-1' }))
expect(result?.status).toBe('working')
expect(result?.result).toBeUndefined()
expect(result?.resultType).toBe('complete')
})
it('returns the stored tool result on completion', async () => {
const storedResult = {
resultType: 'complete',
content: [{ type: 'text', text: '{"file_name":"arkiv.zip"}' }],
structuredContent: { file_name: 'arkiv.zip' },
}
mocks.taskRow = {
id: 'task-1',
company_id: '11111111-1111-4111-8111-111111111111',
user_id: 'user-1',
tool_name: 'gnubok_audit_package',
status: 'completed',
status_message: null,
result: storedResult,
error: null,
poll_interval_ms: 2000,
ttl_ms: 3600000,
created_at: '2026-07-29T09:00:00.000Z',
}
const { result } = await readBody(mcpRequest('tasks/get', { taskId: 'task-1' }))
expect(result?.status).toBe('completed')
expect(result?.result).toEqual(storedResult)
})
it('rejects unknown task ids with invalid params', async () => {
const { error } = await readBody(mcpRequest('tasks/get', { taskId: 'nope' }))
expect(error?.code).toBe(-32602)
})
it('requires taskId', async () => {
const { error } = await readBody(mcpRequest('tasks/get', {}))
expect(error?.code).toBe(-32602)
})
})
describe('tasks/cancel and tasks/update', () => {
it('acknowledges cancellation and flips a working row', async () => {
const { result } = await readBody(mcpRequest('tasks/cancel', { taskId: 'task-1' }))
expect(result?.resultType).toBe('complete')
expect(mocks.taskUpdates).toHaveLength(1)
expect(mocks.taskUpdates[0].status).toBe('cancelled')
})
it('acknowledges tasks/update as a no-op (no input_required flows yet)', async () => {
const { result } = await readBody(
mcpRequest('tasks/update', { taskId: 'task-1', inputResponses: {} })
)
expect(result?.resultType).toBe('complete')
})
})
})
@@ -100,7 +100,7 @@ describe('MCP spec revision 2026-07-28', () => {
])
const capabilities = result?.capabilities as Record<string, unknown>
expect(capabilities.tools).toEqual({ listChanged: false })
expect(capabilities.extensions).toEqual({ 'io.modelcontextprotocol/ui': {} })
expect(capabilities.extensions).toMatchObject({ 'io.modelcontextprotocol/ui': {} })
expect(typeof result?.ttlMs).toBe('number')
expect(result?.cacheScope).toBe('private')
const meta = result?._meta as Record<string, Record<string, unknown>>
@@ -317,7 +317,7 @@ describe('MCP spec revision 2026-07-28', () => {
expect(result?.protocolVersion).toBe('2025-06-18')
expect((result?.serverInfo as Record<string, unknown>).name).toBe('gnubok')
const capabilities = result?.capabilities as Record<string, unknown>
expect(capabilities.extensions).toEqual({ 'io.modelcontextprotocol/ui': {} })
expect(capabilities.extensions).toMatchObject({ 'io.modelcontextprotocol/ui': {} })
})
it('negotiates an initialize requesting 2026-07-28 down to the handshake default', async () => {
+156 -3
View File
@@ -1,4 +1,12 @@
import { NextResponse, after } from 'next/server'
import {
TASKS_EXTENSION_ID,
isTaskCapableClient,
createMcpTask,
resolveMcpTask,
taskToWire,
type McpTaskRow,
} from './tasks'
import {
extractBearerToken,
validateApiKey,
@@ -244,6 +252,11 @@ interface McpTool {
// _meta.ui.resourceUri on the RESULT, so the host renders the widget only when
// asked. (Contrast _meta above, on the definition, which renders on every call.)
uiResourceUri?: string
// Tasks extension: when this predicate returns true for a call from a
// task-capable client, the dispatcher returns a CreateTaskResult and runs
// execute() after the response instead of blocking on it. Not serialized
// into tools/list.
shouldRunAsTask?: (args: Record<string, unknown>) => boolean
execute: (
args: Record<string, unknown>,
companyId: string,
@@ -12234,6 +12247,10 @@ export const tools: McpTool[] = [
idempotentHint: true, // repeat calls produce equivalent archives, fresh URL
openWorldHint: false,
},
// Archive generation is the one genuinely long-running synchronous call
// in the catalog: task-capable clients get a durable handle instead of a
// multi-minute blocking response. Size estimates stay synchronous.
shouldRunAsTask: (args) => args.estimate_only !== true,
async execute(args, companyId, userId, supabase) {
const fiscalPeriodId = args.fiscal_period_id as string
if (!fiscalPeriodId) throw new Error('fiscal_period_id is required')
@@ -15463,9 +15480,13 @@ const SERVER_CAPABILITIES = {
tools: { listChanged: false },
resources: { listChanged: false },
prompts: { listChanged: false },
// MCP Apps (ratified extension): widgets are served as ui:// resources and
// referenced from tool _meta.ui.resourceUri (see widgets/).
extensions: { 'io.modelcontextprotocol/ui': {} },
extensions: {
// MCP Apps (ratified extension): widgets are served as ui:// resources
// and referenced from tool _meta.ui.resourceUri (see widgets/).
'io.modelcontextprotocol/ui': {},
// MCP Tasks: durable handles for long-running tool calls (see tasks.ts).
[TASKS_EXTENSION_ID]: {},
},
}
/**
@@ -15858,6 +15879,9 @@ export async function handleMcpRequest(request: Request): Promise<Response> {
// Revisions are ISO dates, so string comparison orders them correctly.
const statelessClient =
typeof metaVersion === 'string' && metaVersion >= STATELESS_PROTOCOL_VERSION
// Tasks extension: only a client that declared it in THIS request's
// capabilities may ever receive a CreateTaskResult.
const taskCapable = statelessClient && isTaskCapableClient(requestMeta)
// Standard request headers (2026-07-28): when present they must agree with
// the JSON-RPC body. Absence stays accepted: this server supports
@@ -16238,6 +16262,90 @@ export async function handleMcpRequest(request: Request): Promise<Response> {
// mcp.next_hint_followed when the agent's behaviour matches the hint.
checkAndEmitNextHintFollowed(sessionId, toolName, actor, userId, effectiveCompanyId)
// ── Tasks extension (io.modelcontextprotocol/tasks) ──
// Long-running tools return a durable handle immediately to a client
// that declared the extension; the work completes after the response
// (after() keeps the function alive) and the result lands in mcp_tasks
// for tasks/get polling. Runs after every auth/scope/capability guard
// so nothing is ever started for a call that would have been refused.
if (taskCapable && tool.shouldRunAsTask?.(toolArgs)) {
const task = await createMcpTask(supabase, {
companyId: effectiveCompanyId,
userId,
apiKeyId,
toolName,
})
const taskStartedAt = Date.now()
emitAfterResponse(async () => {
try {
const rawResult = await tool.execute(toolArgs, effectiveCompanyId, userId, supabase, actor)
const canonicalResult = addCompanyToTopLevelNext(rawResult, effectiveCompanyId)
const result = projectMcpPayload(canonicalResult, toolNamespace)
const stored: Record<string, unknown> = {
resultType: 'complete',
content: [{ type: 'text', text: JSON.stringify(result, null, 2) }],
}
if (result !== null && result !== undefined) {
stored.structuredContent =
typeof result === 'object' && !Array.isArray(result) ? result : { value: result }
}
await resolveMcpTask(supabase, task.id, { status: 'completed', result: stored })
emitToolCallTelemetry({
tool: toolName,
requiredScope: requiredScope ?? null,
actor,
latencyMs: Date.now() - taskStartedAt,
success: true,
isError: false,
errorCode: null,
errorKind: null,
errorMessage: null,
requestId: id ?? null,
userId,
companyId: effectiveCompanyId,
})
} catch (err) {
// Tool failures complete the task with the standard isError
// envelope: exactly what the synchronous call would have
// returned. `failed` stays reserved for infrastructure errors.
const structured = toToolError(err, { toolName })
const publicStructured = projectMcpPayload(structured, toolNamespace)
await resolveMcpTask(supabase, task.id, {
status: 'completed',
result: {
resultType: 'complete',
content: [{ type: 'text', text: JSON.stringify(publicStructured, null, 2) }],
isError: true,
},
statusMessage: structured.error.message_sv,
}).catch((updateErr) => {
log.error('Failed to store MCP task failure result', { taskId: task.id, updateErr })
})
emitToolCallTelemetry({
tool: toolName,
requiredScope: requiredScope ?? null,
actor,
latencyMs: Date.now() - taskStartedAt,
success: false,
isError: true,
errorCode: structured.error.code,
errorKind: 'execution',
errorMessage: structured.error.message_sv,
requestId: id ?? null,
userId,
companyId: effectiveCompanyId,
})
}
})
return NextResponse.json(
jsonRpc(id ?? null, {
resultType: 'task',
task: taskToWire(task),
_meta: { [META_SERVER_INFO]: SERVER_INFO_BY_NAMESPACE[toolNamespace] },
})
)
}
const callStartedAt = Date.now()
try {
// gnubok_search_tools needs the caller's scopes to filter results to
@@ -16537,6 +16645,51 @@ export async function handleMcpRequest(request: Request): Promise<Response> {
)
}
case 'tasks/get': {
const taskId = (params as Record<string, unknown>)?.taskId
if (typeof taskId !== 'string' || !taskId) {
return NextResponse.json(jsonRpcError(id ?? null, -32602, 'taskId is required'))
}
// Scoped to the creating user: an API key can only poll its own tasks.
const { data: taskRow } = await supabase
.from('mcp_tasks')
.select('*')
.eq('id', taskId)
.eq('user_id', userId)
.single()
if (!taskRow) {
return NextResponse.json(jsonRpcError(id ?? null, -32602, `Task not found: "${taskId}"`))
}
const row = taskRow as McpTaskRow
const wire: Record<string, unknown> = { resultType: 'complete', ...taskToWire(row) }
if (row.status === 'completed' && row.result) wire.result = row.result
if (row.status === 'failed' && row.error) wire.error = row.error
wire._meta = { [META_SERVER_INFO]: SERVER_INFO_BY_NAMESPACE[toolNamespace] }
return NextResponse.json(jsonRpc(id ?? null, wire))
}
case 'tasks/update':
// No input_required flows exist yet: acknowledge and ignore unknown or
// already-satisfied inputResponses, as the extension spec instructs.
return NextResponse.json(jsonRpc(id ?? null, { resultType: 'complete' }))
case 'tasks/cancel': {
const taskId = (params as Record<string, unknown>)?.taskId
if (typeof taskId !== 'string' || !taskId) {
return NextResponse.json(jsonRpcError(id ?? null, -32602, 'taskId is required'))
}
// Cooperative cancellation: flip a still-working row; an in-flight
// execution is not interrupted, and its late completion becomes a
// no-op against the now-terminal row.
await supabase
.from('mcp_tasks')
.update({ status: 'cancelled' })
.eq('id', taskId)
.eq('user_id', userId)
.eq('status', 'working')
return NextResponse.json(jsonRpc(id ?? null, { resultType: 'complete' }))
}
default:
return NextResponse.json(
jsonRpcError(id ?? null, -32601, `Method not found: "${method}"`)
+121
View File
@@ -0,0 +1,121 @@
import type { SupabaseClient } from '@supabase/supabase-js'
/**
* MCP Tasks extension (io.modelcontextprotocol/tasks).
*
* Durable handles for long-running tool calls: a task-capable client gets a
* CreateTaskResult (resultType: "task") immediately and polls tasks/get until
* a terminal status. Rows live in mcp_tasks (service-role writes only) so
* handles survive disconnects and serverless instance turnover.
*
* Failure mapping: a tool execution failure is stored as a COMPLETED task
* whose result carries the standard isError envelope, because that is exactly
* what the synchronous call would have returned. The `failed` status (and the
* `error` column) is reserved for infrastructure failures where no tool
* result exists.
*/
export const TASKS_EXTENSION_ID = 'io.modelcontextprotocol/tasks'
const DEFAULT_POLL_INTERVAL_MS = 2000
const DEFAULT_TTL_MS = 3_600_000
export interface McpTaskRow {
id: string
company_id: string
user_id: string
tool_name: string
status: 'working' | 'input_required' | 'completed' | 'failed' | 'cancelled'
status_message: string | null
result: Record<string, unknown> | null
error: Record<string, unknown> | null
poll_interval_ms: number
ttl_ms: number
created_at: string
}
/**
* Per the extension spec, a server must never return a task to a client that
* did not declare the extension in this request's capabilities.
*/
export function isTaskCapableClient(requestMeta: Record<string, unknown>): boolean {
const caps = requestMeta['io.modelcontextprotocol/clientCapabilities']
if (!caps || typeof caps !== 'object') return false
const extensions = (caps as Record<string, unknown>).extensions
if (!extensions || typeof extensions !== 'object') return false
return TASKS_EXTENSION_ID in (extensions as Record<string, unknown>)
}
export async function createMcpTask(
supabase: SupabaseClient,
params: { companyId: string; userId: string; apiKeyId?: string | null; toolName: string }
): Promise<McpTaskRow> {
// Storage limitation (GDPR Art. 5(1)(e)): opportunistically purge expired
// rows on every creation so the 1-hour retention is enforced without
// dedicated cron infrastructure (cheap via idx_mcp_tasks_expires).
// Best-effort: a failed sweep must never block the new task.
try {
await supabase.from('mcp_tasks').delete().lt('expires_at', new Date().toISOString())
} catch {
// Ignore: the next creation retries the sweep.
}
const { data, error } = await supabase
.from('mcp_tasks')
.insert({
company_id: params.companyId,
user_id: params.userId,
api_key_id: params.apiKeyId ?? null,
tool_name: params.toolName,
status: 'working',
poll_interval_ms: DEFAULT_POLL_INTERVAL_MS,
ttl_ms: DEFAULT_TTL_MS,
})
.select('*')
.single()
if (error || !data) {
throw new Error(`Failed to create MCP task: ${error?.message ?? 'no row returned'}`)
}
return data as McpTaskRow
}
/**
* Move a still-working task to a terminal state. The status='working' guard
* makes terminal states immutable (spec) and lets a tasks/cancel that raced
* the execution win: the late completion becomes a no-op.
*/
export async function resolveMcpTask(
supabase: SupabaseClient,
taskId: string,
terminal: {
status: 'completed' | 'failed' | 'cancelled'
result?: Record<string, unknown>
error?: Record<string, unknown>
statusMessage?: string
}
): Promise<void> {
// Literal payload (no conditional spreads) so the phantom-column guard can
// resolve every column. Writing null for absent terminal fields is correct:
// the transition sets the complete terminal state.
await supabase
.from('mcp_tasks')
.update({
status: terminal.status,
result: terminal.result ?? null,
error: terminal.error ?? null,
status_message: terminal.statusMessage ?? null,
})
.eq('id', taskId)
.eq('status', 'working')
}
/** Map a row to the wire Task object shared by CreateTaskResult and tasks/get. */
export function taskToWire(row: McpTaskRow): Record<string, unknown> {
return {
taskId: row.id,
status: row.status,
createdAt: row.created_at,
ttlMs: Number(row.ttl_ms),
pollIntervalMs: row.poll_interval_ms,
...(row.status_message ? { statusMessage: row.status_message } : {}),
}
}
+1
View File
@@ -961,6 +961,7 @@ export const ARCHIVE_EXCLUDED_TABLES: Record<string, string> = {
idempotency_keys: 'infrastructure',
inbox_rate_counters: 'infrastructure',
invoice_inbox_items: 'inbox workflow state; the files live in document_attachments',
mcp_tasks: 'MCP task handles: transient tool-call state with a 1-hour TTL',
metered_events: 'billing telemetry',
notification_log: 'notification dedup log',
operations: 'staged-operation workflow state',
@@ -0,0 +1,59 @@
-- Migration: MCP Tasks (io.modelcontextprotocol/tasks extension)
-- Durable handles for long-running MCP tool calls: the tool call returns a
-- task handle immediately and the work completes after the response; clients
-- poll tasks/get until a terminal status. Writes go through the service-role
-- MCP handler only (mirrors pending_operations); only the creating user may
-- read. Retention (1 hour via expires_at) is enforced by an opportunistic
-- sweep in the MCP handler: createMcpTask deletes expired rows on every
-- task creation (GDPR Art. 5(1)(e)).
CREATE TABLE public.mcp_tasks (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
company_id UUID NOT NULL REFERENCES public.companies(id) ON DELETE CASCADE,
user_id UUID NOT NULL REFERENCES auth.users(id) ON DELETE CASCADE,
-- Attribution only (which API key created the task); deliberately no FK so
-- key rotation or deletion never breaks task history.
api_key_id UUID,
tool_name TEXT NOT NULL,
status TEXT NOT NULL DEFAULT 'working' CHECK (status IN (
'working', 'input_required', 'completed', 'failed', 'cancelled'
)),
status_message TEXT,
-- Terminal payloads: `result` holds exactly what the synchronous tool call
-- would have returned (a CallToolResult, including isError envelopes);
-- `error` holds a JSON-RPC error object for infrastructure failures.
result JSONB,
error JSONB,
poll_interval_ms INTEGER NOT NULL DEFAULT 2000,
ttl_ms BIGINT NOT NULL DEFAULT 3600000,
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
updated_at TIMESTAMPTZ NOT NULL DEFAULT now(),
expires_at TIMESTAMPTZ NOT NULL DEFAULT now() + interval '1 hour'
);
-- tasks/get looks up by id + creator; expiry cleanup scans expires_at.
CREATE INDEX idx_mcp_tasks_user ON public.mcp_tasks (user_id, status);
CREATE INDEX idx_mcp_tasks_company ON public.mcp_tasks (company_id);
CREATE INDEX idx_mcp_tasks_expires ON public.mcp_tasks (expires_at);
ALTER TABLE public.mcp_tasks ENABLE ROW LEVEL SECURITY;
-- Reads: the creating user only. tasks/get in the MCP handler scopes to the
-- creator, and task results carry whatever the underlying tool returned;
-- the DB grant must not be broader than that application contract
-- (data minimisation, GDPR Art. 5(1)(c)). Mirrors pending_operations.
CREATE POLICY "mcp_tasks_select_own" ON public.mcp_tasks
FOR SELECT USING (auth.uid() = user_id);
-- No INSERT/UPDATE/DELETE policies: all writes go through the service-role
-- MCP handler (mirrors pending_operations). Rows age out via expires_at.
CREATE TRIGGER mcp_tasks_updated_at
BEFORE UPDATE ON public.mcp_tasks
FOR EACH ROW EXECUTE FUNCTION public.update_updated_at_column();
-- No audit trigger: operational task plumbing, not business records. The
-- underlying tool executions already emit mcp.tool_called telemetry, and any
-- committed bookkeeping effects carry their own audit trail.
NOTIFY pgrst, 'reload schema';
+109
View File
@@ -0,0 +1,109 @@
import { randomUUID } from 'node:crypto'
import { describe, expect, it } from 'vitest'
import { getPool, withUserContext } from './setup'
import { seedCompany, insertAuthUser, insertCompanyMember } from './fixtures'
// pg-real coverage for 20260729094000_create_mcp_tasks.sql: the status CHECK,
// creator-only SELECT RLS, the deliberate absence of authenticated write
// policies (writes are service-role only), and the updated_at trigger.
async function insertTask(params: {
companyId: string
userId: string
status?: string
toolName?: string
}): Promise<string> {
const id = randomUUID()
await getPool().query(
`INSERT INTO public.mcp_tasks (id, company_id, user_id, tool_name, status)
VALUES ($1, $2, $3, $4, $5)`,
[id, params.companyId, params.userId, params.toolName ?? 'gnubok_audit_package', params.status ?? 'working'],
)
return id
}
describe('mcp_tasks', () => {
it('rejects statuses outside the task lifecycle', async () => {
const { companyId, userId } = await seedCompany()
await expect(
insertTask({ companyId, userId, status: 'running' }),
).rejects.toThrow(/mcp_tasks_status_check/)
})
it('only the creating user can read a task (not other members, not other companies)', async () => {
const a = await seedCompany()
const b = await seedCompany()
const taskId = await insertTask({ companyId: a.companyId, userId: a.userId })
const mine = await withUserContext(a.userId, (client) =>
client.query('SELECT id, status FROM public.mcp_tasks WHERE id = $1', [taskId]),
)
expect(mine.rows).toHaveLength(1)
expect(mine.rows[0].status).toBe('working')
// A second member of the SAME company must not see the row: task results
// carry raw tool output, so the grant is creator-only (Art. 5(1)(c)).
const colleagueId = await insertAuthUser()
await insertCompanyMember({ companyId: a.companyId, userId: colleagueId, role: 'member' })
const colleague = await withUserContext(colleagueId, (client) =>
client.query('SELECT id FROM public.mcp_tasks WHERE id = $1', [taskId]),
)
expect(colleague.rows).toHaveLength(0)
const theirs = await withUserContext(b.userId, (client) =>
client.query('SELECT id FROM public.mcp_tasks WHERE id = $1', [taskId]),
)
expect(theirs.rows).toHaveLength(0)
})
it('authenticated users cannot insert, update, or delete tasks (service-role only)', async () => {
const { companyId, userId } = await seedCompany()
const taskId = await insertTask({ companyId, userId })
await expect(
withUserContext(userId, (client) =>
client.query(
`INSERT INTO public.mcp_tasks (company_id, user_id, tool_name)
VALUES ($1, $2, 'gnubok_audit_package')`,
[companyId, userId],
),
),
).rejects.toThrow(/row-level security/)
// UPDATE and DELETE have no policies: RLS silently filters all rows,
// so the statements succeed but affect nothing.
const upd = await withUserContext(userId, (client) =>
client.query(`UPDATE public.mcp_tasks SET status = 'cancelled' WHERE id = $1`, [taskId]),
)
expect(upd.rowCount).toBe(0)
const del = await withUserContext(userId, (client) =>
client.query('DELETE FROM public.mcp_tasks WHERE id = $1', [taskId]),
)
expect(del.rowCount).toBe(0)
})
it('bumps updated_at on status transitions', async () => {
const { companyId, userId } = await seedCompany()
const taskId = await insertTask({ companyId, userId })
const before = await getPool().query(
'SELECT updated_at FROM public.mcp_tasks WHERE id = $1',
[taskId],
)
// clock_timestamp()-based trigger: force a measurable gap.
await getPool().query('SELECT pg_sleep(0.05)')
await getPool().query(
`UPDATE public.mcp_tasks SET status = 'completed', result = '{"ok":true}'::jsonb WHERE id = $1`,
[taskId],
)
const after = await getPool().query(
'SELECT updated_at, status FROM public.mcp_tasks WHERE id = $1',
[taskId],
)
expect(after.rows[0].status).toBe('completed')
expect(new Date(after.rows[0].updated_at).getTime()).toBeGreaterThanOrEqual(
new Date(before.rows[0].updated_at).getTime(),
)
})
})