"Papers With Code" matching MCP servers:

Executes Python code in isolated rootless containers while proxying MCP server tools, reducing context overhead by 95%+ and enabling complex multi-tool workflows through sandboxed code execution.

Enables Claude Code to delegate tasks to OpenAI's Codex CLI (GPT-5.4) with structured execution traces, parallel execution, session persistence, and adversarial code review.

Offloads expensive code reading, editing, and checking tasks to DeepSeek through Claude Code, saving Codex main-thread tokens by 40-60%.

Wraps Claude Code as a server using the Model Context Protocol (MCP), allowing teams to interact with Claude Code through a Slack bot interface.

Executes TypeScript code in a secure sandbox with IPC-bridged access to MCP tools like filesystem operations and network requests.

Enables safe execution of Python code with AI assistance and integration with Model Context Protocol tools.

A Code Mode MCP server for the e2b API, giving agents three tools (search, execute_read, execute_write) to write and run JavaScript functions inside sandboxed environments for API discovery and calls.

Enables AI agents to write and execute Python code in an isolated sandbox that can orchestrate multiple MCP tool calls, reducing context window bloat and improving efficiency for complex workflows.

Composes arbitrary MCP tools using JavaScript, enabling discovery, composition, and execution of other MCP servers through a single 'exec' tool.

Enables running one Codex CLI task at a time in an isolated temporary folder, returning Codex's answer and a manifest of produced files for manual review and selective copying.

An MCP bridge that lets Codex supervise Reasonix as an implementation worker in an isolated Git worktree, ensuring safe delegation with scoped writes and sandboxed execution.

Enables agents to interact with the Logflare API by writing JavaScript functions that run in a Vercel Sandbox, using three tools: search, execute_read, and execute_write.

An MCP server that keeps raw tool output out of the context window by sandboxing and compressing results (about 98% reduction), persisting file edits, git operations, tasks and decisions in SQLite with BM25 retrieval so the agent resumes after compaction, and enforcing routing across 17 platforms via MCP and hooks. It also encourages the model to write scripts that compute answers and log only results instead of reading large data into context.

An MCP server for running Claude Code and the Codex CLI as a pair: Claude drives, and hands self-contained tasks to Codex as background jobs.

Enables AI coding assistants to browse Unreal Python API documentation, search classes and functions, and execute Python code in Unreal Editor.

A server that allows LLMs to run Claude Code with all permissions bypassed automatically, enabling code execution and file editing without permission interruptions.

Context Mode is an MCP server that reduces context window waste by sandboxing data-heavy tools, tracking session state in SQLite, and promoting code-based analysis over raw data reads, achieving up to 98% context savings.

An MCP server that preserves LLM context by intercepting large data outputs and returning only concise summaries or relevant sections. It enables efficient sandboxed code execution, file processing, and documentation indexing across multiple programming languages and authenticated CLIs.

Enables a model to write short Python programs that call a wrapped MCP server's tools as async functions, rather than invoking them one at a time. The program runs in a locked-down Docker sandbox, and every tool call passes through a host-side bridge that enforces an allowlist, schema-validated arguments, call budgets, timeouts and output caps.

Enables agents to accomplish multi-step work by writing a single sandboxed JavaScript program that drives shell, filesystem, HTTP, and every tool of any configured MCP server, returning only a small summary instead of raw JSON. This collapses dozens of advertised tool schemas into one tool, cutting per-request context by roughly 26× while keeping disabled upstream servers reachable.