modelcontextprotocol / modelcontextprotocol/python-sdk
[v2] Expose request cancellation on MCPServer tool Context
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 24.3k
- Forks
- 4k
- Avg merge
- 1d 1h
- Merged PRs (30d)
- 31
Description
What happened?
With mcp==2.1.0, a tool registered through the high-level mcp.server.mcpserver.MCPServer API receives mcp.server.mcpserver.Context. That context does not expose a public cancel_requested event, although the lower-level mcp.shared.context.BaseContext does.
from mcp.server.mcpserver.context import Context as MCPServerContext
from mcp.shared.context import BaseContext
assert not hasattr(MCPServerContext, "cancel_requested")
assert hasattr(BaseContext, "cancel_requested")
The server dispatcher processes notifications/cancelled and holds the event in its per-request dispatch context, but normal tool handlers cannot cooperatively observe it through the supported high-level context API. A handler that owns cleanup-sensitive work (for example, a bounded child process) must either rely only on task interruption or traverse private session/dispatch internals to relay cancellation.
Expected behavior
Expose the current request's cancellation event on the high-level tool context (for example, await ctx.cancel_requested.wait()), or document an equivalent supported API. It should allow a handler to trigger its own cooperative cleanup without using private SDK attributes.
If this is already addressed in a later release, a version pointer would also resolve the report; the reproduction above is against 2.1.0.
Reproduction steps
- Install
mcp==2.1.0. - Register an async tool with
MCPServerthat accepts its injectedContext. - Attempt to access
ctx.cancel_requested; the attribute is absent. - Compare with
BaseContext.cancel_requested, which is present.
SDK version
mcp==2.1.0 (Python 3.12)
Area
Server / high-level MCPServer tool context API.
AI assistance disclosure: drafted with AI assistance; the reporter is responsible for this report.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start at mcp.server.mcpserver.MCPServer and its injected mcp.server.mcpserver.context.Context, then compare how mcp.shared.context.BaseContext exposes cancel_requested and how notifications/cancelled is stored in the per-request dispatch context. Done means high-level tool handlers can use a supported cancellation event without private session or dispatch internals, with coverage for the reported reproduction.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- api, backend
- Issue type
- Feature
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 68/100