MCP Completion Issue

Question

mcp = FastMCP("completion1",
              instructions="A server that generates code from user prompts and JSON context. For all validation code generation requests from developers")

# Global store for extracted values
repo_cache = []

@mcp.tool()
async def long_task(files: list[str], ctx: Context) -> str:
    """Process a list of files asynchronously.

    Args:
        files (list[str]): _description_
        ctx (Context): _description_

    Returns:
        str: _description_
    """
    global repo_cache
    repo_cache.clear()  # reset

    for i, file in enumerate(files):
        ctx.info(f"Processing {file}")
        with open(file, "r", encoding="utf-8") as f:
            data = f.read()

        # Just for illustration: suppose lines in file are repo names
        lines = data.splitlines()
        repo_cache.extend(lines)

        await ctx.report_progress(i + 1, len(files))

    ctx.info(f"Cached repos: {repo_cache}")
    return "Processing complete"


@mcp.completion()
async def free_text_completion(argument: CompletionArgument) -> Completion:
    """Handle free text completion requests.
    Args:
        argument (CompletionArgument): The argument containing the user's input.
    Returns:
        Completion: A completion object containing suggestions based on the user's input.
    """    
    # Use the global repo_cache to provide suggestions
    global repo_cache
    suggestions = repo_cache or ["code_genrator", "some_other_repo"]
    matches = [r for r in suggestions if r.startswith(argument.value)]
    return Completion(values=matches)


if __name__ == "__main__":
    mcp.run(transport="stdio")

we have three queries as follows-
I have implemented a context loader & a completion endpoint as above I'm using mcp tools & it's completion endpoint with GitHub copilot in VS code.

Hear are my queries-

How can I get my recommendation using completion endpoint in completion1.
Where I will be getting the recommendation in VS code file editor or in GitHub copilot while MCP is connected & is running.
What are the additional implementations I can add here to make the MCP'c workflow better.
My mcp for now is loading the context & further generation are happening through it one I have sent a query from GitHub copilot.

Please help with the queries I have written above.

I too have faced the same issues when I was working with the MCPs implementation for my environment. I struggled finding a proper implementation of MCP completion with GitHub copilot in VS Code.

I’m facing the same issue

MCP completions apply to exactly two things — the arguments of a prompt and the parameters of a resource template — and the client sends completion/complete while its user is filling one of those in (see the Completions guide). A server that only registers a tool never receives that request, so free_text_completion is never called: tool arguments are chosen by the model, not through completions. To answer (1), give the handler something to complete — e.g. a prompt with a repo argument — and use the full (ref, argument, context) handler signature; the example below adapts your repo cache and drives it with the in-memory client (v2 API: FastMCP is now MCPServer). For (2), where and whether those suggestions appear is the client's UI decision (VS Code / Copilot), not something this SDK controls. For (3), to get the cached repo list in front of the model, expose it as a resource or return it from a tool — completions are for a human filling in a form.

Example

import anyio
from mcp_types import Completion, CompletionArgument, CompletionContext, PromptReference, ResourceTemplateReference

from mcp import Client
from mcp.server import MCPServer

mcp = MCPServer("completion1")
repo_cache = ["code_generator", "some_other_repo"]


@mcp.prompt()
def generate(repo: str, request: str) -> str:
    """Generate validation code for a repo."""
    return f"Generate {request} for repo {repo}"


@mcp.completion()
async def suggest_repo(
    ref: PromptReference | ResourceTemplateReference,
    argument: CompletionArgument,
    context: CompletionContext | None,
) -> Completion | None:
    if isinstance(ref, PromptReference) and ref.name == "generate" and argument.name == "repo":
        return Completion(values=[r for r in repo_cache if r.startswith(argument.value)])
    return None


async def main() -> None:
    async with Client(mcp) as client:
        result = await client.complete(
            ref=PromptReference(name="generate"),
            argument={"name": "repo", "value": "code"},
        )
        print(result.completion.values)  # ['code_generator']


anyio.run(main)

Closing this as part of a wider backlog cleanup around the v2 release. Feel free to reopen if this is still relevant.

_{AI Disclaimer}

Navigation Menu

Repository navigation

MCP Completion Issue #1065

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

MCP Completion Issue #1065

Description

Activity

Amitt1412 commented on Jul 1, 2025

akshada-pawar22 commented on Jul 1, 2025

maxisbey commented on Jul 24, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions