Case · DiggingBeagle record

MCP Python SDK tasks crossed client-session boundaries

An opt-in experimental tasks feature in the MCP Python SDK used task identifiers without checking which session created them, allowing other clients to inspect or cancel tasks.

CVE-2026-52870 applies only to applications that enabled the low-level experimental tasks feature. It does not describe every MCP Python SDK deployment.

First seen
Jun 5, 2026
Case kind
vulnerability
AI role
AGAINST AI
Claims
3

Reconstruction

The defect was ordinary authorization inside an agent protocol feature. Experimental task handlers accepted a task ID as enough authority for list, get, result and cancel operations. In a multi-client server, the SDK did not record and enforce which session owned the task.

That matters because asynchronous agent work often contains long-lived results, prompts and elicitation state. A task identifier can be guessed, discovered or leaked; it should not become a bearer credential unless the protocol explicitly designs it that way.

The advisory is tightly scoped. Applications that never enabled the experimental tasks feature were not affected. Version 1.27.2 is listed as the fix.

Mechanism & boundary

  1. 01

    Client creates a task

    Experimental tasks create a long-lived task identifier.

    Boundary: client / task store

  2. 02

    Server stores the task without creator-session ownership

    Affected handlers can later resolve the task by identifier alone.

    Boundary: task identity / authorization

  3. 03

    Another client uses task operations

    A different connected client can list, read results from or cancel the task.

    Boundary: session A / session B

Timeline

  1. Jun 5, 2026

    GitHub advisory published

    report

    The MCP Python SDK advisory documents the experimental-task authorization flaw.

  2. Jul 16, 2026

    Advisory metadata updated

    followup

    The reviewed advisory record was updated in July.

Claims & evidence

reported findingsupported

The default experimental task handlers did not bind a task to the session that created it, so another connected client could observe, read results from or cancel that task.

reported findingsupported

CVE-2026-52870 affected MCP Python SDK versions 1.23.0 through 1.27.1 when applications explicitly enabled the experimental tasks feature.

Evidence visuals

diagram

MCP task ownership gap

  1. Client A

    Creates an experimental task

  2. Task store

    Task ID stored without creator-session ownership check

  3. Client B

    Uses list/get/result/cancel against another client's task

  • Client A Task store: create task
  • Client B Task store: cross-client operations
Project-authored reconstruction from GHSA-hvrp-rf83-w775. · Source: MCP Python SDK tasks crossed client-session boundaries

Implications

Agent protocols need object-level authorization for asynchronous work. Session ownership should be explicit at creation and enforced at every subsequent operation.

Controls & mitigations

  • Upgrade affected deployments to mcp 1.27.2 or later.
  • Bind every task to an authenticated session or principal at creation.
  • Authorize list, get, result, cancel and related elicitation operations against that ownership.

What remains unknown

  • The public advisory does not establish exploitation in the wild.

Cite this record

DiggingBeagle. “MCP Python SDK tasks crossed client-session boundaries.” First seen Jun 5, 2026. https://diggingbeagle.com/cases/mcp-python-sdk-tasks-crossed-client-session-boundaries/

Citation guidance

Why this archive exists

The source matters after the headline fades.

DiggingBeagle is a non profit research project documenting AI security incidents, agent failures, vulnerabilities and AI-assisted operations. A case keeps its claims beside the sources that support, contest or limit them. Later updates stay visible, so a reader can see when the account changed.

We publish case reconstructions, dated reporting and analysis across records. Each has a different evidentiary role. About the project and our methodology explain how the work is reviewed.