Add `Agent.to_mcp()` method #3076

Kludex · 2025-10-03T09:19:30Z

No description provided.

cursor

This is the final PR Bugbot will review for you during this billing cycle

Your free Bugbot reviews will reset on November 9

Details

You are on the Bugbot Free tier. On this plan, Bugbot will review limited PRs each billing cycle.

To receive Bugbot reviews on all of your PRs, visit the Cursor dashboard to activate Pro and start your 14-day free trial.

Bug: Image Data Decoding Error

The map_from_pai_messages function incorrectly calls base64.b64decode() on image chunk.data. Since chunk.data is already binary, not a base64 string, this causes a decoding error and prevents proper image content conversion for MCP.

pydantic_ai_slim/pydantic_ai/_mcp.py#L93-L95

pydantic-ai/pydantic_ai_slim/pydantic_ai/_mcp.py

Lines 93 to 95 in 0f5e987

    
           if isinstance(chunk, str): 
        
               add_msg('user', mcp_types.TextContent(type='text', text=chunk)) 
        
           elif isinstance(chunk, messages.BinaryContent) and chunk.is_image:

pydantic_ai_slim/pydantic_ai/_mcp.py

github-actions · 2025-10-03T09:26:56Z

Docs Preview

commit:	`0f5e987`
Preview URL:	https://67f184e5-pydantic-ai-previews.pydantic.workers.dev

DouweM · 2025-10-03T14:00:53Z

pydantic_ai_slim/pydantic_ai/_mcp.py

+    server_name: str | None = None,
+    tool_name: str | None = None,
+    tool_description: str | None = None,
+    # TODO(Marcelo): Should this actually be a factory that is created in every tool call?


I think a union of static deps and a deps factory makes sense, if the deps factory would get the tool call _meta.

DouweM · 2025-10-03T14:14:21Z

pydantic_ai_slim/pydantic_ai/_mcp.py

+                name=tool_name,
+                description=tool_description,
+                inputSchema={'type': 'object', 'properties': {'prompt': {'type': 'string'}}},
+                # TODO(Marcelo): How do I get this?


There's not currently a nice way to get this, but it'd be useful to have a new output_json_schema property on AbstractAgent.

In the case of the concrete Agent, it would depend on agent._output_schema:

if StructuredTextOutputSchema (superclass of NativeOutputSchema and PromptedOutputSchema), get it from .object_def.json_schema

if it's OutputSchemaWithoutMode, get it from .processor.object_def.json_schema

if it's PlainTextOutputSchema, it's just {'type': 'string'}

if it's ToolOutputSchema, we need to create a union schema of all .toolset.processors using UnionOutputProcessor, which currently takes a sequence of output types and creates ObjectOutputProcessors for them on the fly, but could also take a sequence of ObjectOutputProcessors (the ones we get from .toolset.processors) directly. Once we have that UnionOutputProcessor, the schema is on .object_def.json_schema

if it's ToolOrTextOutputSchema, it's the above or {'type': 'string'}

Note that this changes a bit with some refactoring I did in #2970, but directionally it's the same: there's not currently a nice way to get this, and it's especially tricky for tool output, because we don't have a union of all types handy.

I should be able to implement this pretty quickly through, once that images PR with the output types refactor merges.

Should this PR wait for it then?

I'll wait for it then.

DouweM · 2025-10-03T14:15:00Z

pydantic_ai_slim/pydantic_ai/_mcp.py

+            raise ValueError(f'Unknown tool: {name}')
+
+        # TODO(Marcelo): Should we pass the `message_history` instead?
+        prompt = cast(str, args['prompt'])


Can we use a typed dict for args so we don't have to cast?

No, that would be incorrect... What I actually need to check if 'prompt' is in args, and check that it's a string.

I'd expect the library to validate the args match the type hint, no?

Explained via Slack - answering here: no.

DouweM · 2025-10-03T14:16:01Z

pydantic_ai_slim/pydantic_ai/_mcp.py

+        if name != tool_name:
+            raise ValueError(f'Unknown tool: {name}')
+
+        # TODO(Marcelo): Should we pass the `message_history` instead?


I think just the prompt is fine, when would the LLM generate an entire message history?

Hmmm, I think the point is that we need to maintain the history in the session...

Good point!

We may need to create a database abstraction here. 🤔

Are you sure the tool should be stateful like that? If it's essentially a subagent, wouldn't multiple calls be expected to start separate subagents? I think continuing the conversation should be explicit, with some conversation ID returned and passed in

Yes, if the client wants to create a new conversation, they can open a new session.

I think continuing the conversation should be explicit, with some conversation ID returned and passed in

The MCP spec handles this with a session ID.

DouweM · 2025-10-03T14:16:10Z

pydantic_ai_slim/pydantic_ai/_mcp.py

+
+        return dict(result=result.output)
+
+    app.list_tools()(list_tools)


These could be decorators right?

Decorators inside a function are treated as misused type-wise.

Kludex · 2025-10-06T09:31:16Z

PR pending work on #3076 (comment)

Add Agent.to_mcp() method

0f5e987

cursor bot reviewed Oct 3, 2025

View reviewed changes

pydantic_ai_slim/pydantic_ai/_mcp.py Show resolved Hide resolved

DouweM requested changes Oct 3, 2025

View reviewed changes

Kludex marked this pull request as draft October 6, 2025 09:30

	if isinstance(chunk, str):
	add_msg('user', mcp_types.TextContent(type='text', text=chunk))
	elif isinstance(chunk, messages.BinaryContent) and chunk.is_image:


		return dict(result=result.output)

		app.list_tools()(list_tools)

Add Agent.to_mcp() method #3076

Are you sure you want to change the base?

Add Agent.to_mcp() method #3076

Conversation

Kludex commented Oct 3, 2025

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

This is the final PR Bugbot will review for you during this billing cycle

Bug: Image Data Decoding Error

Uh oh!

Uh oh!

github-actions bot commented Oct 3, 2025

Docs Preview

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Kludex commented Oct 6, 2025

Uh oh!

Uh oh!

Add `Agent.to_mcp()` method #3076

Add `Agent.to_mcp()` method #3076