You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
New integration page: OnlineCompiler.io (code execution and test-case judging over MCP) #751
Hi! We'd like to add an integration page for OnlineCompiler.io, in the same form as the Unstructured Transform MCP and Prior Labs pages: documentation only, using mcp-haystack's MCPToolset, with no new package.
What it gives a Haystack agent
run_code: runs one program in any of 46 languages (Python, Java, C++, Go, Rust, SQL and others) with optional stdin, and returns stdout, stderr, exit code, time and memory.
run_tests: runs a program against input and expected-output pairs and returns a verdict for each test, an overall verdict and a score.
list_compilers, and judge_problem for problems stored in an OnlineCompiler.io account.
Each run happens in a fresh gVisor sandbox with no network access. The hosted server is at https://api.onlinecompiler.io/mcp (bearer token), and the same server runs locally over stdio from PyPI (onlinecompiler-mcp).
Where it fits next to E2B
E2B gives the agent a sandbox session that it manages over several steps. This covers the single-shot cases: run one program and read its output, or judge a solution against test cases, which tutoring, grading and code-checking agents need.
What the PR adds
integrations/onlinecompiler.md (type Tool Integration, mcp: true; pypi points at mcp-haystack, repo at our MIT-licensed MCP server)
logos/onlinecompiler.png
The examples are tested with haystack-ai 3.5.0 and mcp-haystack 2.1.2. I've opened the PR alongside so you can see the page; happy to change anything. We maintain the MCP server and will keep the page current.
Activity