Repository navigation
Missing prompt-level _meta support in MCPServer.prompt() and Prompt model #3476
Description
Initial Checks
- I confirm that I'm using the newest release of my line (the latest 2.x, or the latest 1.x if I'm still on v1)
- I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue
Release line
2.x (current stable)
Description
Description
The 2026-07-28 specification defines the optional _meta field on the Prompt schema. While other decorators like @mcp.tool support meta and pass it down as _meta on serialization (e.g., in MCPServer.list_tools). The @mcp.prompt decorator and the high-level Prompt classes are missing support for meta.
Expected Behavior
Including meta on @mcp.prompt does not incur a type error and returns _meta for the prompt on list_prompts.
Example Code
from mcp.server.mcpserver import MCPServer
# ATTEMPTING TO USE METADATA ON PROMPTS
mcp = MCPServer("MetaTestServer")
# This currently raises a TypeError because meta is not a parameter in @mcp.prompt
@mcp.prompt(name="code_review",title="Code Review",
description="Review code with specific metadata rules",
meta={"strictness": "high"})
def review_code(code: str) -> str:
return f"Review this: {code}"
# 2. EXPECTED BEHAVIOR
# If the decorator parameter was supported, list_prompts should return prompt with _meta:
# listed_prompts = await mcp.list_prompts()
# assert listed_prompts[0].meta == {"audience": "strictness": "high"}Python & MCP Python SDK
- Python: 3.12+
- MCP Python SDK: 2.2.0
Activity
Ben-Home commentedon Sep 8, 2026
feat(mcpserver): add prompt-level meta support to MCPServer.prompt an…
feat(mcpserver): add prompt-level meta support to MCPServer.prompt an…
siddharthgaur1 commentedon Sep 8, 2026
I've opened #3479 for this. It was auto-closed because I'm not assigned here — the bot says it reopens on its own if a maintainer assigns me, so this is a request for that rather than anything needing action on the PR.
The wire type already supports it: on
main,Prompt(name="x", _meta={"a": 1}).model_dump(by_alias=True)gives{'name': 'x', '_meta': {'a': 1}}. Only the high-level layer was missing it, so the change mirrors what@mcp.tooland@mcp.resourcealready do — ametafield on the high-levelPrompt, ametaargument onPrompt.from_functionand the decorator, and_meta=prompt.metainlist_promptsnext to the existing_meta=info.metainlist_tools. Three source lines added, one changed.Three tests next to the existing
TestServerResourceMetadata, including your example verbatim; two of them fail onmainwith theTypeErrorfrom the issue.tests/server/mcpserver/: 633 passed, 1 skipped.ruffclean, andpyrighton the changed files shows only the pre-existingasynccontextmanagerdeprecation atserver.py:149.I left
get_promptalone deliberately — it doesn't carry meta into the protocol response, mirroringread_resource, and this issue asks only aboutlist_prompts. Happy to extend it if you'd rather have both.Written with AI assistance (Claude Code); I reproduced the
TypeErroronmain@9972c21and read the tool/resource meta paths before mirroring them.