Even though the 'tools/call' request to 'notification_tool' is successful and an 'MCP::Tool::Response' is returned, the logger still prints 'Error:'.
```
% ruby examples/streamable_http_client.rb
=== MCP SSE Test Client ===
[CLIENT] INFO 15:41:04.696 - Initializing session...
[CLIENT] INFO 15:41:04.698 - Session initialized: SESSION_ID
[CLIENT] INFO 15:41:04.698 - Server info: {"name" => "sse_test_server", "version" => "0.1.0"}
[CLIENT] INFO 15:41:04.698 - Connecting to SSE stream...
[CLIENT] INFO 15:41:04.698 - SSE stream connected successfully
=== Available Actions ===
1. Send custom notification
2. Test echo
3. List tools
0. Exit
Choose an action: 1
Enter notification message: notification test
Enter delay in seconds (0 for immediate):
[CLIENT] INFO 15:41:11.248 - SSE data: {"jsonrpc":"2.0","id":"0f657eeb-3e05-47bb-b7b4-898e2cb503af","result":{"content":[{"type":"text","text":"Notification: notification test (timestamp: 2025-06-21T15:41:11+09:00)"}],"isError":false}}
[CLIENT] ERROR 15:41:11.249 - Error:
=== Available Actions ===
1. Send custom notification
2. Test echo
3. List tools
0. Exit
```
It is probably trying to output an MCP::Tool::Response, but these appear in the SSE stream. Therefore, the response it is handling this time should either be accepted or an error.
I modified it so that it could handle them.
This was causing conflicts in applications using `json-schema` 5.x.
The only breaking change in 5.0 was dropping support for old Ruby
versions, which we don't support either.
This PR renamed the files related to the SSE example and fixed the README.md.
Because missing files were specified in README.md.
And also, I removed the word 'test' from the file names related to the SSE example.
Other examples file names do not contain the word `test`.
`handle_request` already adds the `method` to the instrumentation data,
so there is no need to repeat this. Moreover, consumers can override
these anyway.
The reason for the change is that a NameError occurs when a POST request is made while the stream is closed because ModelContextProtocol is undefined.
I have also added test cases to check the behaviour when IOError and Errno::EPIPE occurs.
Follow-up to #27.
This PR adds a migration path for `Server::Transports::StdioTransport`. e.g.,
```console
$ cat /tmp/example.rb
require "mcp/transports/stdio"
MCP::Transports::StdioTransport
```
This change displays the following warning instead of raising an error for such cases, allowing execution to proceed:
```console
$ bundle exec ruby /tmp/example.rb
/tmp/example.rb:1: warning: Use `require "mcp/server/transports/stdio_transport"` instead of `require "mcp/transports/stdio"`.
Also use `MCP::Server::Transports::StdioTransport` instead of `MCP::Transports::StdioTransport`.
This API is deprecated and will be removed in a future release.
```
Currently (v0.1.0), it results in an error without any warning.
This MCP gem is not yet close to version 1.0, but introducing breaking changes to a published interface is undesirable.
This compatibility code should be removed at some point after a few releases.
Closes#50.
This PR adds links to the following two badges:
1. RubyGems badge: A link to the `mcp` gem will be added. The repository name is `ruby-sdk`,
following the modelcontextprotocol org's naming convention. Since the repository name and
the gem name are different, the link will help make it clear which gem it refers to.
2. MIT license badge: This will link to the MIT license file in the repository.
NOTE: Separately from this PR, the copyright notice in the MIT license should probably be addressed.
It currently says `Copyright (c) 2025 TODO: Write your name`, so the TODO should be resolved:
https://github.com/modelcontextprotocol/ruby-sdk/blob/v0.1.0/LICENSE.txt#L3
Other language SDKs seem to list `Anthropic, PBC` as the copyright holder,
though I would defer that decision to someone more knowledgeable.
This PR suppresses the following backtrace of `Interrupt` (Ctrl+c) exception for stdio server:
## Before
```console
$ ./examples/stdio_server.rb
^C/Users/koic/src/github.com/modelcontextprotocol/ruby-sdk/lib/mcp/server/transports/stdio_transport.rb:20:
in 'IO#gets': Interrupt
from /Users/koic/src/github.com/modelcontextprotocol/ruby-sdk/lib/mcp/server/transports/stdio_transport.rb:20:
in 'MCP::Server::Transports::StdioTransport#open'
from ./examples/stdio_server.rb:95:in '<main>'
$ echo $?
130
```
## After
```console
$ ./examples/stdio_server.rb
^C
Exiting...
$ echo $?
130
```
This change should prevent unexpected backtrace from appearing when running the example:
https://github.com/modelcontextprotocol/ruby-sdk?tab=readme-ov-file#stdio-transport
`MCP::Resource` doesn't have `text` attribute:
https://github.com/modelcontextprotocol/ruby-sdk/blob/v0.1.0/lib/mcp/resource.rb#L8-L13
I've replaced it with attributes that it actually supports.
Also, the use of `text: "Hello, world!"` in the example of `server.resources_read_handler` implementation resulted in
an unexpected response the first time I worked with `Resource`. While it may not be the best (but better),
in my experience using Resources with Cline, `text: params[:uri]` served as a clearer response when interacting with an LLM,
making it easier to understand.
Refactor capabilities handling into MCP::Server::Capabilities to improve
code organization and separation of concerns. This provides a cleaner API
for configuring server capabilities including prompts, resources, tools,
completions, logging, and experimental features.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
