diff --git a/README.md b/README.md
index c70eb57..5487d1f 100644
--- a/README.md
+++ b/README.md
@@ -29,10 +29,11 @@ Phase 1 runtime is implemented:
 Phase 2 (MCP surface, `workplans/QONTO-WP-0003-mcp-surface.md`) is in
 progress: a streamable-HTTP MCP adapter is mounted at `/mcp` on the same
 capability core and policy kernel as REST, with tools `qonto_org_summary`,
-`qonto_list_transactions`, and `qonto_cost_run_rate_hints`. Client auth is
-still the REST-style `X-Actor-*` header convention; real OIDC/workload auth,
-the shared multi-harness client config snippet, and the `finance-qonto-read`
-tool profile are not yet implemented.
+`qonto_list_transactions`, and `qonto_cost_run_rate_hints`. The endpoint is
+gated by a shared-secret bearer token (`QONTO_ASSISTANT_MCP_TOKEN`) — see
+`docs/mcp-integration.md` for the auth model, its gap vs. the OIDC/workload
+target, and the shared multi-harness client config snippet. The
+`finance-qonto-read` tool profile is not yet implemented.
 
 Current verification:
 
diff --git a/docs/mcp-integration.md b/docs/mcp-integration.md
new file mode 100644
index 0000000..15d905d
--- /dev/null
+++ b/docs/mcp-integration.md
@@ -0,0 +1,119 @@
+# MCP Integration (Phase 2)
+
+`qonto-assistant` exposes a streamable-HTTP MCP adapter at `/mcp`, mounted on
+the same FastAPI process as the REST surface. It shares the exact policy
+kernel and audit layer REST uses — see `specs/ArchitectureBlueprint.md` §6.
+
+Tools exposed (read-only, default-deny policy underneath):
+
+| MCP tool | Capability | Notes |
+| --- | --- | --- |
+| `qonto_ping` | none (no bank call) | Smoke/liveness check |
+| `qonto_org_summary` | `org_summary` | Organization + per-account balances |
+| `qonto_list_transactions` | `list_transactions` | Capped, filtered history |
+| `qonto_cost_run_rate_hints` | `cost_run_rate_hints` | Normalized recurring-cost hints, not a raw export |
+
+No spend, transfer, card, or volume-cost tool is ever exposed here — those
+stay hard-denied by the shared policy kernel regardless of transport.
+
+## Auth model
+
+Two independent layers, same as the blueprint's identity separation
+(`specs/ArchitectureBlueprint.md` §4.7):
+
+1. **Workload auth (is this caller allowed to reach the service at all?)**
+   Today: a static shared-secret bearer token
+   (`QONTO_ASSISTANT_MCP_TOKEN`), checked by `BearerTokenAuthMiddleware`
+   in front of `/mcp`. If the env var is unset, the middleware is not
+   installed — used only for local fixture-backed smoke work, never for a
+   deployment holding real credentials.
+
+   This is **not** the OIDC/workload-identity target the blueprint
+   describes — there is no OIDC issuer in this fleet yet to federate
+   against. A shared-secret bearer token is the deployable "or similar"
+   primitive for now; upgrading to real OIDC/workload identity is
+   fleet-level follow-on work (tracked as a Phase 2 gap, not closed by this
+   task).
+
+   This token is a **service credential**, not a bank credential — it never
+   reaches Qonto and is never logged. Treat it like any other shared
+   secret: store it in OpenBao or your harness's secret manager, not in
+   plaintext config committed to a repo.
+
+2. **Actor identity (who is calling, for policy and audit purposes?)**
+   Same `X-Actor-ID` / `X-Tenant-ID` / `X-Actor-Lane` / `X-Actor-Scopes`
+   header convention REST already uses (`auth.actor_claims_from_headers`).
+   These are self-asserted today, not cryptographically bound to the bearer
+   token — tightening that binding is exactly what Phase 3's flex-auth
+   resource (`finance.qonto.read`) is for.
+
+## One shared client config snippet
+
+The same `url` + `headers` shape works across MCP-capable harnesses that
+support a remote streamable-HTTP server (Claude Code, Claude Desktop,
+Cursor, and Codex/Grok-style harnesses that follow the MCP client config
+convention):
+
+```json
+{
+  "mcpServers": {
+    "qonto-assistant": {
+      "url": "http://127.0.0.1:8080/mcp",
+      "headers": {
+        "Authorization": "Bearer ${QONTO_ASSISTANT_MCP_TOKEN}",
+        "X-Actor-ID": "${AGENT_ACTOR_ID}",
+        "X-Tenant-ID": "binky"
+      }
+    }
+  }
+}
+```
+
+Replace the `url` host/port with wherever the service is actually deployed.
+`${QONTO_ASSISTANT_MCP_TOKEN}` and `${AGENT_ACTOR_ID}` are placeholders for
+whatever secret-interpolation syntax your harness config supports — never
+paste the literal token into a shared config file.
+
+## Local smoke: start with auth enabled, connect with only this snippet
+
+```bash
+export QONTO_FIXTURE_DIR=tests/fixtures/qonto
+export QONTO_ASSISTANT_MCP_TOKEN=dev-local-token
+make run   # or: PYTHONPATH=src ../state-hub/.venv/bin/python -m qonto_assistant.main
+```
+
+Then, using only the URL and `Authorization: Bearer dev-local-token` header
+from the snippet above, connect with the `mcp` Python SDK's client (or any
+MCP Inspector-style tool):
+
+```python
+import asyncio
+from mcp import ClientSession
+from mcp.client.streamable_http import streamablehttp_client
+
+async def main():
+    async with streamablehttp_client(
+        "http://127.0.0.1:8080/mcp",
+        headers={
+            "Authorization": "Bearer dev-local-token",
+            "X-Actor-ID": "local-operator",
+            "X-Tenant-ID": "binky",
+        },
+    ) as (read, write, _):
+        async with ClientSession(read, write) as session:
+            await session.initialize()
+            print([tool.name for tool in (await session.list_tools()).tools])
+            print((await session.call_tool("qonto_org_summary", {})).structuredContent)
+
+asyncio.run(main())
+```
+
+A request without the `Authorization` header (or with the wrong token) gets
+a `401 {"error_code": "unauthorized"}` before it ever reaches a tool.
+
+## Related
+
+- `docs/operator-runbook.md` — REST operator runbook (credential sources,
+  fixture mode, `/v1/*` endpoints)
+- `specs/ArchitectureBlueprint.md` §6 Phase 2 — the plan this doc implements
+- `workplans/QONTO-WP-0003-mcp-surface.md` — Phase 2 workplan and task status
diff --git a/docs/operator-runbook.md b/docs/operator-runbook.md
index 5318555..4c20435 100644
--- a/docs/operator-runbook.md
+++ b/docs/operator-runbook.md
@@ -162,6 +162,13 @@ PY
 Use `window_days=31` for a recent-activity view. Use `window_days=90` or `93`
 when recurring fixed-cost hints are required.
 
+## MCP surface
+
+Phase 2 mounts a streamable-HTTP MCP adapter at `/mcp` on this same process,
+sharing the policy kernel and audit layer above. See
+`docs/mcp-integration.md` for the tool catalog, the auth model
+(`QONTO_ASSISTANT_MCP_TOKEN`), and the shared client config snippet.
+
 ## CostRunRate refresh path
 
 `binky-control` should consume `GET /v1/snapshot` and extract:
diff --git a/src/qonto_assistant/app.py b/src/qonto_assistant/app.py
index 7bfac9e..bdc5d07 100644
--- a/src/qonto_assistant/app.py
+++ b/src/qonto_assistant/app.py
@@ -16,6 +16,7 @@ from qonto_assistant.auth import actor_claims_from_request
 from qonto_assistant.config import Settings
 from qonto_assistant.credentials import build_credential_provider
 from qonto_assistant.errors import QontoAssistantError, UpstreamError
+from qonto_assistant.mcp_auth import BearerTokenAuthMiddleware
 from qonto_assistant.mcp_server import create_mcp_server
 from qonto_assistant.policy import PolicyEngine
 from qonto_assistant.qonto_client import FixtureQontoClient, QontoClient
@@ -42,6 +43,8 @@ def create_app(
 
     mcp_server = create_mcp_server(settings=settings, service=service)
     mcp_app = mcp_server.streamable_http_app()
+    if settings.mcp_auth_token:
+        mcp_app.add_middleware(BearerTokenAuthMiddleware, token=settings.mcp_auth_token)
 
     @asynccontextmanager
     async def lifespan(_: FastAPI) -> AsyncIterator[None]:
diff --git a/src/qonto_assistant/config.py b/src/qonto_assistant/config.py
index 53770a6..f391dfd 100644
--- a/src/qonto_assistant/config.py
+++ b/src/qonto_assistant/config.py
@@ -32,6 +32,7 @@ class Settings:
     openbao_path: str
     openbao_command: str
     openbao_timeout_seconds: float
+    mcp_auth_token: str | None
     host: str
     port: int
 
@@ -69,6 +70,7 @@ class Settings:
             openbao_path=os.getenv("QONTO_OPENBAO_PATH", "tenants/binky/qonto-api"),
             openbao_command=os.getenv("QONTO_OPENBAO_COMMAND", "bao"),
             openbao_timeout_seconds=float(os.getenv("QONTO_OPENBAO_TIMEOUT_SECONDS", "5")),
+            mcp_auth_token=os.getenv("QONTO_ASSISTANT_MCP_TOKEN") or None,
             host=os.getenv("QONTO_ASSISTANT_HOST", "127.0.0.1"),
             port=int(os.getenv("QONTO_ASSISTANT_PORT", "8080")),
         )
diff --git a/src/qonto_assistant/mcp_auth.py b/src/qonto_assistant/mcp_auth.py
new file mode 100644
index 0000000..0b04ca3
--- /dev/null
+++ b/src/qonto_assistant/mcp_auth.py
@@ -0,0 +1,52 @@
+from __future__ import annotations
+
+import hmac
+
+from starlette.requests import Headers
+from starlette.responses import JSONResponse
+from starlette.types import ASGIApp, Receive, Scope, Send
+
+
+class BearerTokenAuthMiddleware:
+    """Workload auth gate for the MCP endpoint: static shared-secret bearer token.
+
+    QONTO-WP-0003-T03 scope note: the blueprint's target is OIDC/workload
+    identity (specs/ArchitectureBlueprint.md §4.7), but no OIDC issuer exists
+    in this fleet yet. A shared-secret bearer token is the "or similar"
+    workload-auth primitive that's actually deployable today -- it keeps
+    unauthenticated callers off the MCP surface entirely, same custody
+    principle as REST: this token is a service credential, never a bank
+    credential, and per-actor identity still comes from X-Actor-* headers
+    checked by the policy kernel. Upgrading to real OIDC/workload identity
+    federation is fleet-level follow-on work, not something this repo can
+    stand up alone.
+    """
+
+    def __init__(self, app: ASGIApp, *, token: str) -> None:
+        self.app = app
+        self.token = token
+
+    async def __call__(self, scope: Scope, receive: Receive, send: Send) -> None:
+        if scope["type"] != "http":
+            await self.app(scope, receive, send)
+            return
+
+        headers = Headers(scope=scope)
+        if not _token_matches(headers.get("authorization", ""), self.token):
+            response = JSONResponse(
+                {"error_code": "unauthorized", "detail": "Missing or invalid bearer token"},
+                status_code=401,
+                headers={"WWW-Authenticate": "Bearer"},
+            )
+            await response(scope, receive, send)
+            return
+
+        await self.app(scope, receive, send)
+
+
+def _token_matches(authorization_header: str, expected_token: str) -> bool:
+    prefix = "Bearer "
+    if not authorization_header.startswith(prefix):
+        return False
+    presented = authorization_header[len(prefix) :]
+    return hmac.compare_digest(presented, expected_token)
diff --git a/tests/test_api.py b/tests/test_api.py
index 1f99315..8f3b2fe 100644
--- a/tests/test_api.py
+++ b/tests/test_api.py
@@ -38,6 +38,7 @@ def _settings() -> Settings:
         openbao_path="tenants/binky/qonto-api",
         openbao_command="bao",
         openbao_timeout_seconds=5,
+        mcp_auth_token=None,
         host="127.0.0.1",
         port=8080,
     )
diff --git a/tests/test_mcp_auth.py b/tests/test_mcp_auth.py
new file mode 100644
index 0000000..e55609d
--- /dev/null
+++ b/tests/test_mcp_auth.py
@@ -0,0 +1,35 @@
+from starlette.applications import Starlette
+from starlette.responses import PlainTextResponse
+from starlette.routing import Route
+from starlette.testclient import TestClient
+
+from qonto_assistant.mcp_auth import BearerTokenAuthMiddleware
+
+
+def _protected_app() -> Starlette:
+    async def ok(request):
+        return PlainTextResponse("ok")
+
+    app = Starlette(routes=[Route("/", ok)])
+    app.add_middleware(BearerTokenAuthMiddleware, token="secret-token")
+    return app
+
+
+def test_bearer_auth_rejects_missing_header() -> None:
+    client = TestClient(_protected_app())
+    response = client.get("/")
+    assert response.status_code == 401
+    assert response.json()["error_code"] == "unauthorized"
+
+
+def test_bearer_auth_rejects_wrong_token() -> None:
+    client = TestClient(_protected_app())
+    response = client.get("/", headers={"Authorization": "Bearer wrong-token"})
+    assert response.status_code == 401
+
+
+def test_bearer_auth_accepts_matching_token() -> None:
+    client = TestClient(_protected_app())
+    response = client.get("/", headers={"Authorization": "Bearer secret-token"})
+    assert response.status_code == 200
+    assert response.text == "ok"
diff --git a/workplans/QONTO-WP-0003-mcp-surface.md b/workplans/QONTO-WP-0003-mcp-surface.md
index 90018b7..6a58bd9 100644
--- a/workplans/QONTO-WP-0003-mcp-surface.md
+++ b/workplans/QONTO-WP-0003-mcp-surface.md
@@ -122,7 +122,7 @@ REST. `pytest` → `25 passed`; `python3 -m compileall src tests scripts`.
 
 ```task
 id: QONTO-WP-0003-T03
-status: todo
+status: done
 priority: high
 state_hub_task_id: "ad43b09a-ffa5-4df0-8999-bfcef1f11732"
 ```
@@ -137,6 +137,40 @@ Done when: the snippet is in `docs/operator-runbook.md` (or a new
 `docs/mcp-integration.md`), and a local smoke connects a real MCP client
 against the running adapter using only that snippet.
 
+**Done 2026-07-23 — with a scoped gap, called out explicitly:** No OIDC
+issuer exists anywhere in this fleet yet, so standing up real
+OIDC/workload-identity federation isn't something this repo can do alone —
+implementing it would mean pointing `FastMCP`'s OAuth Protected Resource
+flow at an issuer URL that doesn't serve real metadata, which is worse than
+not having it. Shipped the deployable primitive instead: a shared-secret
+bearer token (`QONTO_ASSISTANT_MCP_TOKEN`), enforced by
+`src/qonto_assistant/mcp_auth.py::BearerTokenAuthMiddleware` in front of
+`/mcp` only (REST untouched). Unset by default (fixture/local smoke stays
+credential-free); when set, every non-matching or missing
+`Authorization: Bearer <token>` gets `401 {"error_code": "unauthorized"}`
+before reaching any tool or the policy kernel. Constant-time comparison
+(`hmac.compare_digest`). This token is a **service credential, not a bank
+credential** — never logged, never reaches Qonto. Per-actor identity is
+still the self-asserted `X-Actor-*` header convention (not yet
+cryptographically bound to the bearer token — that binding is Phase 3's
+flex-auth resource `finance.qonto.read`, not this task).
+
+Added `docs/mcp-integration.md`: tool table, the two-layer auth model
+explained (workload auth today vs. the OIDC target and why it's deferred),
+and the one shared `{"mcpServers": {...}}` config snippet (`url` + `headers`
+— the convention shared by Claude Code, Claude Desktop, Cursor, and
+Codex/Grok-style harnesses for remote streamable-HTTP MCP servers).
+
+Verified live, exactly as the doc's smoke section describes: started the
+service with `QONTO_ASSISTANT_MCP_TOKEN` set and `QONTO_FIXTURE_DIR` (no
+real Qonto credentials); a request with no `Authorization` header never
+reached a tool; a request with the wrong token got `401 Unauthorized` in the
+server log; a request built from *only* the doc's snippet (URL + the two
+headers) listed tools and called `qonto_org_summary` successfully, with
+`X-Actor-ID` flowing into the audit event as before. New
+`tests/test_mcp_auth.py` covers missing/wrong/matching token cases directly.
+`pytest` → `28 passed`; `python3 -m compileall src tests scripts`.
+
 ## Task: agent-harness tool profile `finance-qonto-read`
 
 ```task