feat: route mcp runtime prompt gets

This commit is contained in:
lda
2026-06-07 19:13:23 +07:00 Verified
parent f89801e7a8
commit b6bdae2020
7 changed files with 130 additions and 4 deletions
@@ -0,0 +1,438 @@
# MCP Runtime Get Prompt Implementation Plan
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
**Goal:** Add persistent MCP runtime support for `get_prompt()` by routing it through the existing owner-task operation queue and `McpSourceClient`.
**Architecture:** This is the second safe non-tool runtime operation after `read_resource`. It stays a thin wrapper over `McpSourceClient.get_prompt()` and returns serialized `dict[str, Any]`. Raw method invocation, notifications, and discovery lists remain separate future decisions.
**Tech Stack:** Python 3.14, MCP Python SDK, pytest/pytest-asyncio, ruff, basedpyright.
---
## File Structure
- Modify `src/wf_sources_mcp/runtime/factory.py`
- Add `_SessionOwner.get_prompt(prompt_name, arguments=None)`.
- Implement through `submit(operation="get_prompt", run=lambda client: client.get_prompt(prompt_name, arguments))`.
- Modify `src/wf_sources_mcp/runtime/session.py`
- Add `RawPromptGetter` callback type.
- Add `get_prompt_callback` field.
- Add `get_prompt(prompt_name, arguments=None)` method.
- Keep fallback `client.get_prompt()` serialization for injected/fake sessions.
- Modify `src/wf_sources_mcp/runtime/pool.py`
- Add `McpRuntimePool.get_prompt(connection, auth, prompt_name, arguments=None)`.
- Modify tests:
- `tests/wf_sources_mcp/test_runtime.py`
- `tests/wf_mcp/test_stateful_runtime.py` only if type compatibility requires it.
- Update docs:
- `docs/current_roadmap.md`
- `docs/superpowers/specs/2026-06-03-long-lived-workflow-api-boundary.md`
## Hard Boundaries
- Do not add persistent runtime `invoke_method`, `send_notification`, or discovery list methods.
- Do not expose raw `ClientSession` outside the owner task.
- Do not return raw MCP SDK result models from `get_prompt`; return serialized `dict[str, Any]`.
- Do not alter `McpSourceClient.get_prompt()` behavior.
- Do not dispatch by string; `operation="get_prompt"` is metadata only.
---
### Task 1: Add Persistent Prompt Tests
**Files:**
- Modify: `tests/wf_sources_mcp/test_runtime.py`
- [ ] **Step 1: Extend fake client with prompt gets**
In `tests/wf_sources_mcp/test_runtime.py`, update `_FakeFactory._create_with_stack()`'s `_FakeClient` class to include:
```python
async def get_prompt(
self,
prompt_name: str,
arguments: dict[str, str] | None = None,
):
return type(
"GetPromptResult",
(),
{
"model_dump": lambda _self, **_kwargs: {
"messages": [
{
"role": "user",
"content": {
"type": "text",
"text": f"{prompt_name}:{arguments or {}}",
},
}
]
}
},
)()
```
- [ ] **Step 2: Add error-path test**
Append to `tests/wf_sources_mcp/test_runtime.py`:
```python
@pytest.mark.asyncio
async def test_persistent_session_raises_without_prompt_transport() -> None:
session = PersistentMcpSession(connection=_connection(), auth=None)
with pytest.raises(RuntimeError, match="no prompt transport"):
await session.get_prompt("prompt.summarize")
```
- [ ] **Step 3: Add session-level persistent prompt test**
Append:
```python
@pytest.mark.asyncio
async def test_persistent_session_factory_routes_prompts_through_owner() -> None:
factory = _FakeFactory()
connection = _connection()
session = await factory.create(connection, None)
await session.call_tool("echo", {"text": "one"})
await session.read_resource("fixture://docs/welcome")
prompt_payload = await session.get_prompt(
"prompt.summarize",
{"text": "hello"},
)
await session.close()
assert factory.created_connections == [connection]
assert factory.calls == [("echo", {"text": "one"})]
assert prompt_payload["messages"][0]["content"]["text"] == (
"prompt.summarize:{'text': 'hello'}"
)
```
- [ ] **Step 4: Add pool-level persistent prompt test**
Append:
```python
@pytest.mark.asyncio
async def test_runtime_pool_reuses_session_for_tool_resource_and_prompt() -> None:
factory = _FakeFactory()
pool = McpRuntimePool(factory.create)
connection = _connection()
tool_result = await pool.call_tool(connection, None, "echo", {"text": "one"})
resource_payload = await pool.read_resource(
connection,
None,
"fixture://docs/welcome",
)
prompt_payload = await pool.get_prompt(
connection,
None,
"prompt.summarize",
{"text": "hello"},
)
await pool.close_all()
assert tool_result.output == {"echoed": "one"}
assert resource_payload["contents"][0]["text"] == "resource text"
assert prompt_payload["messages"][0]["content"]["text"] == (
"prompt.summarize:{'text': 'hello'}"
)
assert factory.created_connections == [connection]
```
- [ ] **Step 5: Update public surface test**
Update `test_persistent_session_public_runtime_exposes_only_tool_and_resource_read` to allow `get_prompt` and still forbid the rest:
```python
def test_persistent_session_public_runtime_exposes_safe_read_operations() -> None:
public_operations = {
name
for name in dir(PersistentMcpSession)
if not name.startswith("_") and callable(getattr(PersistentMcpSession, name))
}
assert "call_tool" in public_operations
assert "read_resource" in public_operations
assert "get_prompt" in public_operations
assert "invoke_method" not in public_operations
assert "send_notification" not in public_operations
```
- [ ] **Step 6: Run tests and confirm failure**
Run:
```bash
uv run pytest tests/wf_sources_mcp/test_runtime.py -q
```
Expected: fail because `PersistentMcpSession.get_prompt` and `McpRuntimePool.get_prompt` do not exist yet.
---
### Task 2: Add `get_prompt` to Persistent Session
**Files:**
- Modify: `src/wf_sources_mcp/runtime/session.py`
- [ ] **Step 1: Add callback type**
Add near existing callback types:
```python
RawPromptGetter = Callable[
[str, dict[str, str] | None],
Awaitable[dict[str, Any]],
]
```
- [ ] **Step 2: Add callback field**
In `PersistentMcpSession`, add:
```python
get_prompt_callback: RawPromptGetter | None = None
```
- [ ] **Step 3: Add `get_prompt` method**
Add below `read_resource()`:
```python
async def get_prompt(
self,
prompt_name: str,
arguments: dict[str, str] | None = None,
) -> dict[str, Any]:
"""Get an MCP prompt through the owner task or injected session."""
if self.get_prompt_callback is not None:
return await self.get_prompt_callback(prompt_name, arguments)
if self.client is not None:
result = await self.client.get_prompt(prompt_name, arguments)
return result.model_dump(by_alias=True, mode="json", exclude_none=True)
raise RuntimeError("persistent MCP session has no prompt transport")
```
- [ ] **Step 4: Run source tests**
Run:
```bash
uv run pytest tests/wf_sources_mcp/test_runtime.py -q
```
Expected: still fail at factory/pool routing until callbacks are wired.
---
### Task 3: Route `get_prompt` Through Owner Queue
**Files:**
- Modify: `src/wf_sources_mcp/runtime/factory.py`
- [ ] **Step 1: Wire callback in `PersistentSessionFactory.create()`**
Update returned `PersistentMcpSession`:
```python
return PersistentMcpSession(
connection=connection,
auth=auth,
call_callback=owner.call_tool,
read_resource_callback=owner.read_resource,
get_prompt_callback=owner.get_prompt,
close_callback=owner.close,
)
```
- [ ] **Step 2: Add owner `get_prompt()`**
Add below `_SessionOwner.read_resource()`:
```python
async def get_prompt(
self,
prompt_name: str,
arguments: dict[str, str] | None = None,
) -> dict[str, Any]:
"""Submit a prompt get through the generic owner-task operation queue."""
return await self.submit(
operation="get_prompt",
run=lambda client: client.get_prompt(prompt_name, arguments),
)
```
- [ ] **Step 3: Run source tests**
Run:
```bash
uv run pytest tests/wf_sources_mcp/test_runtime.py -q
```
Expected: pool test still fails until pool method is added; direct session prompt should pass.
---
### Task 4: Add `get_prompt` to Runtime Pool
**Files:**
- Modify: `src/wf_sources_mcp/runtime/pool.py`
- [ ] **Step 1: Add pool method**
Add below `read_resource()`:
```python
async def get_prompt(
self,
connection: McpSourceConnection,
auth: AuthRecord | None,
prompt_name: str,
arguments: dict[str, str] | None = None,
) -> dict[str, Any]:
session = await self.get_session(connection, auth)
return await session.get_prompt(prompt_name, arguments)
```
- [ ] **Step 2: Run runtime tests**
Run:
```bash
uv run pytest tests/wf_sources_mcp/test_runtime.py tests/wf_mcp/test_stateful_runtime.py -q
```
Expected: pass.
---
### Task 5: Verify Surface and Update Docs
**Files:**
- Modify: `docs/current_roadmap.md`
- Modify: `docs/superpowers/specs/2026-06-03-long-lived-workflow-api-boundary.md`
- Move after completion: `docs/superpowers/plans/2026-06-07-mcp-runtime-get-prompt.md` to `docs/historical/superpowers/plans/2026-06-07-mcp-runtime-get-prompt.md`
- [ ] **Step 1: Verify forbidden runtime methods remain absent**
Run:
```bash
rg -n "def (list_tools|list_resources|list_prompts|invoke_method|send_notification)" src/wf_sources_mcp/runtime
```
Expected: no matches. `read_resource` and `get_prompt` are now allowed.
- [ ] **Step 2: Update roadmap**
In `docs/current_roadmap.md`, under the MCP upstream source runtime cleanup bullets, add:
```markdown
- Completed: persistent MCP runtime can now route `get_prompt` through
the owner-task queue and `McpSourceClient`. This keeps prompt reads
stateful without adding raw method invocation or notification support.
```
- [ ] **Step 3: Update long-lived API boundary spec**
In `docs/superpowers/specs/2026-06-03-long-lived-workflow-api-boundary.md`, add after the `read_resource` runtime item:
```markdown
12. Complete: persistent MCP runtime can route `get_prompt` through the
owner-task queue and `McpSourceClient`. Runtime still does not expose raw
method invocation, notifications, or discovery list operations.
```
Renumber following item if needed.
- [ ] **Step 4: Archive completed plan after implementation**
Run only after all code/tests pass:
```bash
git mv docs/superpowers/plans/2026-06-07-mcp-runtime-get-prompt.md docs/historical/superpowers/plans/2026-06-07-mcp-runtime-get-prompt.md
```
---
### Task 6: Final Verification
**Files:**
- All changed files.
- [ ] **Step 1: Run focused tests**
Run:
```bash
uv run pytest tests/wf_sources_mcp tests/wf_mcp/test_stateful_runtime.py tests/wf_mcp/test_compat_imports.py -q
```
Expected: pass.
- [ ] **Step 2: Run source typecheck**
Run:
```bash
uv run basedpyright --level error src
```
Expected: `0 errors, 0 warnings, 0 notes`.
- [ ] **Step 3: Run focused test typecheck**
Run:
```bash
uv run basedpyright --level error tests/wf_sources_mcp/test_runtime.py tests/wf_mcp/test_stateful_runtime.py
```
Expected: `0 errors, 0 warnings, 0 notes`.
- [ ] **Step 4: Run focused lint**
Run:
```bash
uv run ruff check src/wf_sources_mcp/runtime tests/wf_sources_mcp/test_runtime.py tests/wf_mcp/test_stateful_runtime.py
```
Expected: `All checks passed!`
- [ ] **Step 5: Confirm no forbidden runtime expansion**
Run:
```bash
rg -n "def (list_tools|list_resources|list_prompts|invoke_method|send_notification)" src/wf_sources_mcp/runtime
```
Expected: no matches.
- [ ] **Step 6: Check whitespace**
Run:
```bash
git diff --check
```
Expected: no whitespace errors. CRLF warnings are acceptable on Windows.
---
## Self-Review
- Spec coverage: The plan adds persistent prompt gets through the existing generic queue and source-client facade while keeping the surface narrow.
- Placeholder scan: No placeholder steps remain.
- Type consistency: `get_prompt` returns `dict[str, Any]` everywhere and accepts `dict[str, str] | None` arguments.
- Hard boundary: no raw method, notification, or discovery-list runtime methods are added.