Packages

FastMCP 4.2.3: Breaking Changes, OAuth & OpenTelemetry Security Fixes, and Production Upgrade Guide

FastMCP 4.2.3 ships with opt-in parent_task_spans tracing and OpenAPI path validation fixes, capping a week of rapid 4.2.x updates that dropped Python 3.10 support and patched critical OAuth and OpenTelemetry ReDoS flaws. With 64.5 million monthly PyPI downloads, upgrading safely requires checking your Python runtime floor, OAuth consent behavior, and deprecated 4.1 skills APIs. Here is the exact changelog breakdown, ecosystem comparison against the official MCP Python SDK 2.3.0, and a step-by-step upgrade checklist.

By 0xScorPionSon

9 min read

FastMCP 4.2.3: Breaking Changes, OAuth & OpenTelemetry Security Fixes, and Production Upgrade Guide

FastMCP 4.2.3 Architecture showing Python MCP tool nodes, SkillCatalogProvider, and OpenTelemetry security shield

Executive Summary

  • Package & Ecosystem: fastmcp (Python / PyPI), maintained by Prefect under the Apache-2.0 license.
  • Latest Verified Release: v4.2.3 ("Trace Elements"), published on November 19, 2026 (GitHub Releases), following rapid-fire v4.2.2 (November 18), v4.2.1 (November 17), and v4.2.0 (November 13) releases. All package metadata and download figures in this guide were verified on November 21, 2026.
  • AI & MCP Relevance: FastMCP is the most widely adopted high-level Python framework for building Model Context Protocol (MCP) servers, clients, and interactive AI agent tools, recording 64,575,308 downloads over the last month (PyPI Stats).
  • Primary Upgrade Takeaway: Upgrading to v4.2.3 brings opt-in parent_task_spans distributed tracing, critical OAuth and OpenTelemetry ReDoS (CVE-2026-126360) security patches, and OpenAPI validation fixes—but enforces a Python 3.11+ runtime floor, deprecates the 4.1 skills API ahead of v4.3, and requires upgrading opentelemetry-sdk alongside fastmcp if you use pip (FastMCP Changelog).

Adoption Telemetry & Version Changelog

FastMCP has evolved from a single-file wrapper into the dominant application layer for Python-based MCP infrastructure. While FastMCP 1.0 was originally folded into the official MCP Python SDK in 2024, the standalone fastmcp package continued independent development and now builds directly on top of the official mcp v2 protocol layer (eSEOspace).

As of November 21, 2026, verified ecosystem telemetry shows massive production velocity:

  • Daily PyPI Downloads: 2,866,631 (PyPI Stats)
  • Weekly PyPI Downloads: 17,681,484 (PyPI Stats)
  • Monthly PyPI Downloads: 64,575,308 (PyPI Stats)
  • GitHub Repository Scale: 28.7k stars, 2.6k forks, and 410+ contributors (FastMCP Homepage)
  • Distribution Footprint (v4.2.3): 8.2 kB pure-Python wheel (fastmcp-4.2.3-py3-none-any.whl), 41.5 MB source distribution (fastmcp-4.2.3.tar.gz), uploaded via uv/0.13.9 using PyPI Trusted Publishing (PyPI). Note that PyPI currently reports "No provenance" attestation metadata on the 4.2.3 release files, whereas the underlying mcp 2.3.0 package ships with Full provenance (PyPI mcp).

The Seven-Day 4.2.x Release Cadence

Between November 13 and November 19, 2026, the FastMCP maintainers shipped four consecutive releases that reshaped skills delivery, security defaults, and observability:

Version Codename Release Date Key Milestone & Impact Source
v4.2.0 Mad Skills Nov 13, 2026 Drops Python 3.10 (>=3.11 required); adds SkillCatalogProvider, SkillsDirectoryProvider, SkillManifestProvider, Annotated Depends(), and CallArgument. GitHub
v4.2.1 4, 2, 1... Liftoff Nov 17, 2026 Security & correctness patch: rejects future nbf JWTs, pins opentelemetry-api>=1.45.2 (CVE-2026-126360), updates dnspython to v2.9.2, hardens OAuth proxy. GitHub
v4.2.2 Tea for Two (Too) Nov 18, 2026 Fixes skills/list discovery returning empty for tool-only clients when ResourcesAsTools transform is active. GitHub
v4.2.3 Trace Elements Nov 19, 2026 Adds opt-in parent_task_spans for background tasks; fixes 3 output schema conversion bugs; raises ValidationError on missing OpenAPI path args; requires pydocket>=0.29.1 in fastmcp-tasks. GitHub

What Actually Changed in v4.2.3

In distributed agentic workloads, long-running MCP tool calls frequently execute asynchronously via the fastmcp-tasks extension (@mcp.tool(task=True)). Prior to v4.2.3, background worker spans always initiated their own independent OpenTelemetry trace linked back to the originating request. With v4.2.3, engineers can enable parent_task_spans so that each worker's span nests directly inside the tools/call request span that submitted it—producing a single unified trace across the request and its asynchronous execution (GitHub Releases). It remains off by default to preserve backwards-compatible trace structures.

v4.2.3 also closes three subtle JSON Schema edge cases in output schema conversion: ignoring non-object defaults during object default merging (#6552), resolving recursive $ref pointers inside unpinned positional arrays (#6553), and skipping placeholder validation for missing prefix positions in short positional arrays (#6572). Additionally, calling an auto-generated OpenAPI tool without a required path parameter now raises a clean ValidationError (#6559) instead of bubbling up an unhandled server error, and fastmcp-tasks bumps its minimum dependency floor to pydocket>=0.29.1 (GitHub Releases).


Security Patches, Breaking Changes & Internals

For production teams running FastMCP in multi-tenant or internet-facing environments, v4.2.1 through v4.2.3 are essential security upgrades, not routine feature bumps.

1. OpenTelemetry traceparent ReDoS (CVE-2026-126360)

FastMCP v4.2.1 raised its opentelemetry-api dependency floor to >=1.45.2 (#6513) (FastMCP pyproject.toml). This directly mitigates CVE-2026-126360, a quadratic-time Regular Expression Denial of Service (ReDoS) vulnerability in TraceContextTextMapPropagator.extract() (GitLab Advisory).

While standard HTTP headers forbid embedded newlines, security advisories from Snyk and GitLab explicitly warn that non-HTTP carriers such as JSON bodies—including MCP request metadata in params._meta—can preserve embedded newlines and trigger catastrophic regex backtracking on an async server's event-loop thread (Snyk). A single malformed traceparent value can stall unrelated requests sharing the same event loop.

Crucial pip caveat: FastMCP's maintainers warn that if you upgrade via pip and already have opentelemetry-sdk installed, you must upgrade both packages together (pip install -U fastmcp opentelemetry-sdk) to avoid version-skew conflicts (FastMCP Changelog).

2. OAuth Proxy, JWT Validation, and Supply-Chain Hardening

Across v4.2.0 and v4.2.1, FastMCP tightened several authentication and sandboxing boundaries:

  • Future nbf JWT Rejection (#6512): Tokens whose nbf (Not Before) claim is in the future are now strictly rejected (GitHub Releases).
  • OAuth Consent Binding (#6530): When running with require_authorization_consent="remember", consent is now scoped per redirect URI and scope set. Existing users will be prompted for consent once more per client after upgrading (FastMCP Changelog).
  • Issuer Pinning & Scope Preservation (#6535, #6536): Pre-registered OAuth clients can be pinned to their issuer, and provider required_scopes are preserved even when a custom token_verifier is supplied (GitHub Releases).
  • Transitive Dependency Floors: v4.2.1 bumped dnspython to v2.9.2 for security (#6506), while v4.2.0 enforced patched Authlib and joserfc releases, added wall-clock timeouts and callback limits to MontySandboxProvider and CodeMode (#6063), and included DNS resolution inside SSRF deadlines (#5509) (GitHub Releases).

3. Breaking Changes & Updated Code Patterns

Before bumping your lockfile from 4.1.x or 3.x to 4.2.3, account for three breaking shifts introduced in v4.2.0:

  1. Python 3.11+ Required (#6377): Python 3.10 support is dropped. CI runners, Docker base images (python:3.11-slim or newer), and pyproject.toml constraints must be updated (GitHub Releases).
  2. 4.1 Skills API Deprecated (#6280): Legacy providers under fastmcp.server.providers.skills and helpers in fastmcp.utilities.skills emit deprecation warnings and are scheduled for removal in v4.3. Authoring now uses Skill, SkillCatalogProvider, SkillsDirectoryProvider, or SkillManifestProvider from fastmcp.skills.provider, while wire payloads use MCPSkill (GitHub Releases).
  3. Annotated Dependency Injection (#6378, #6438): Dependencies can now live inside Python type annotations using Annotated[dict, Depends(get_config)] alongside CallArgument, keeping runtime dependencies out of the JSON Schema exposed to LLMs (GitHub Releases).

Here is a clean v4.2.3 server example combining SkillCatalogProvider, Annotated dependency injection, and ResourcesAsTools compatibility (which v4.2.2 fixed for tool-only MCP clients):

from typing import Annotated
from fastmcp import FastMCP, Skill
from fastmcp.dependencies import Depends
from fastmcp.skills.provider import SkillCatalogProvider

def get_runtime_config() -> dict[str, str]:
    return {"env": "production", "region": "us-east-1"}

ConfigDep = Annotated[dict[str, str], Depends(get_runtime_config)]

deploy_skill = Skill.from_directory("skills/deploy")
mcp = FastMCP("Production Agent Server")
mcp.add_provider(SkillCatalogProvider([deploy_skill]))

@mcp.tool
def check_cluster_status(service: str, config: ConfigDep) -> dict[str, str]:
    """Check service health using injected runtime configuration."""
    return {
        "service": service,
        "environment": config["env"],
        "region": config["region"],
        "status": "healthy",
    }

if __name__ == "__main__":
    mcp.run(transport="streamable-http")

Notice that config never appears in the tool's MCP input schema—the LLM only sees service: str, while FastMCP injects get_runtime_config() on the server side (FastMCP Updates).


Ecosystem Comparison & Upgrade Checklist

A common architectural question in late 2026 is whether to build on standalone fastmcp 4.2.3 or directly on the official mcp 2.3.0 Python SDK (MCPServer). Because FastMCP 4 depends on the mcp v2 SDK under the hood, both speak the 2026-07-28 stateless MCP specification while negotiating compatibility with legacy clients (eSEOspace).

Dimension fastmcp (v4.2.3) Official mcp SDK (v2.3.0) Source
Maintainer Prefect (PrefectHQ/fastmcp) Model Context Protocol (modelcontextprotocol/python-sdk) PyPI fastmcp, PyPI mcp
License Apache-2.0 MIT PyPI fastmcp, PyPI mcp
Latest Release 4.2.3 (Nov 19, 2026) 2.3.0 (Oct 2, 2026) GitHub, PyPI mcp
Python Floor >=3.11 >=3.10 (PyPI metadata; README lists 3.11+) PyPI fastmcp, PyPI mcp
PyPI Provenance No provenance on 4.2.3 Full provenance attestation PyPI fastmcp, PyPI mcp
Background Tasks Supported via fastmcp-tasks (pydocket>=0.29.1) + parent_task_spans Not implemented yet in v2 notes GitHub, eSEOspace
Skills & Composition SkillCatalogProvider, server mounting, proxying, middleware, ClientGroup Minimal reference MCPServer + Client GitHub, eSEOspace

When to choose mcp 2.3.0: Pick the official Tier 1 SDK if your company strictly requires MIT licensing, PyPI build provenance attestations, minimal dependencies, or a simple single-server wrapper with a few tools (eSEOspace).

When to choose fastmcp 4.2.3: Pick FastMCP if you need server composition, OpenAPI/FastAPI auto-wrapping, SkillCatalogProvider distribution, ClientGroup multi-server routing, built-in OAuth proxy hardening, or asynchronous background tasks (fastmcp-tasks) (GitHub Releases).

Safe Upgrade & Pinning Checklist

  1. Verify Python Runtime: Confirm all deployment containers and virtual environments run Python 3.11 or newer (python --version).
  2. Upgrade with Co-Dependencies:
    • Using uv (recommended by Prefect): uv add "fastmcp>=4.2.3"
    • Using pip with OpenTelemetry installed: pip install -U "fastmcp>=4.2.3" "opentelemetry-sdk>=1.45.2" (FastMCP Changelog)
    • If using background tasks: verify fastmcp-tasks resolves pydocket>=0.29.1 (GitHub Releases).
  3. Migrate 4.1 Skills Imports: Replace fastmcp.server.providers.skills with from fastmcp.skills.provider import SkillsDirectoryProvider or SkillCatalogProvider, and update wire-model checks from Skill to MCPSkill before v4.3 removes the legacy paths (GitHub Releases).
  4. Audit OAuth Consent & Traces: Expect one-time re-consent prompts for clients using require_authorization_consent="remember", and decide whether to enable parent_task_spans=True if your observability stack expects nested worker spans (GitHub Releases).

Frequently Asked Questions (FAQ)

Does FastMCP 4.2.3 still support Python 3.10?

No. Starting with v4.2.0 (November 13, 2026), FastMCP raised its minimum requirement to Python >=3.11 (PyPI). Teams stuck on Python 3.10 must either stay pinned to the 4.1.x branch or use the official mcp 2.3.0 SDK, which still lists >=3.10 in its PyPI metadata (PyPI mcp).

Why do I need to upgrade opentelemetry-sdk alongside fastmcp?

FastMCP v4.2.1 raised its opentelemetry-api floor to >=1.45.2 to patch CVE-2026-126360, a quadratic-time ReDoS flaw in traceparent parsing that can affect JSON-carried MCP metadata (params._meta) (Snyk). Because pip may not automatically bump a pre-installed opentelemetry-sdk to match opentelemetry-api, Prefect explicitly advises running pip install -U fastmcp opentelemetry-sdk (FastMCP Changelog).

Is FastMCP 4 compatible with the official MCP Python SDK v2?

Yes. FastMCP 4 is built directly on top of the official mcp v2 SDK and supports the 2026-07-28 stateless MCP specification while negotiating backwards compatibility for legacy clients (eSEOspace). However, if you are migrating code from mcp.server.fastmcp (SDK v1), note that SDK v2 renamed its built-in class to MCPServer and removed the mcp.server.fastmcp import path (PyPI mcp).

What happens if I don't migrate my 4.1 skills code right away?

In v4.2.x, the old fastmcp.server.providers.skills and fastmcp.utilities.skills modules still function and emit deprecation warnings, but Prefect has scheduled them for hard removal in v4.3 (GitHub Releases). Pin <4.3 in your dependency constraints if you need extra time to migrate to fastmcp.skills.provider.

Will parent_task_spans break my existing OpenTelemetry dashboards?

No, because parent_task_spans is opt-in and disabled by default in v4.2.3 (GitHub Releases). Existing deployments keep separate linked worker traces unless you explicitly enable it to nest worker spans under the submitting tools/call span.


Sources

More on Packages

Packages

OpenAI @openai/codex 0.171.0: GPT-6.2 Sol Default, TUI Workspace, and Sandbox Hardening

OpenAI released @openai/codex 0.171.0 on November 20, 2026, switching the default CLI model to GPT-6.2 Sol, enabling the TUI workspace with document and image previews by default, and adding remote app-server execution. Backed by 23.2 million weekly npm downloads and a Rust 1.99.1 core, the release also hardens Linux seccomp sandboxing and enterprise MCP sign-in.

8 min read

Packages

openai-agents 0.24.0–0.25.0 vs. @openai/agents 0.22.0: Breaking Changes, MCP v2, and Sandbox Trade-Offs

OpenAI's Agents SDK ships as two fast-moving pre-1.0 packages: openai-agents on PyPI (0.24.0 stable / 0.25.0 release files) and @openai/agents on npm (0.22.0, MIT). Because minor releases introduce breaking RunState schemas, MCP approval-mode renames, and strict sandbox write policies, pinning exact versions is essential. Here is the side-by-side changelog, dependency footprint, and safe upgrade guide.

6 min read

Packages

agent-framework v1.26.0 vs. Malicious 1.26.1–1.26.2: PyPI Compromise Telemetry and Safe Pinning Guide

Microsoft's agent-framework PyPI package (193,470 monthly downloads) suffered a brief supply-chain compromise on November 17, 2026, when unauthorized 1.26.1 and 1.26.2 wheels dropped a Python .pth startup stealer targeting OPENAI, HF_, AZURE, and GitHub Actions OIDC tokens. PyPI removed both releases, leaving v1.26.0 (November 12, 2026) as the latest clean version. Here is the exact telemetry, SHA-256 verification, and safe pinning checklist.

6 min read

Tell us what is not working

A thirty-minute call, no deck and no obligation. You will leave with a straight answer about whether this is worth building - including if that answer is no.

Start a project

Direct line

hello@codebumble.net
Dhaka, Bangladesh  :   local timeCalls in English or Bangla