Roundtable MCP Server for IDE Model Delegation

Overview

Roundtable is a local Model Context Protocol (MCP) server that lets your primary IDE assistant delegate subtasks to specialized AI models. Instead of manually switching between different chat windows or copying context across tools, you send one high-level prompt from your IDE; Roundtable fans out that workload to multiple sub-agents (e.g., Gemini, Claude, Codex, Cursor), keeps a single shared project context, runs the agents in parallel, and returns synthesized results back to the primary assistant.

This approach preserves context continuity, reduces wait time by parallelizing analysis and code generation, and allows you to use the most appropriate model for each task. Roundtable runs locally, invokes the CLI tools you already have installed and configured, and does not add any markup or billing overhead — it uses your existing API keys and subscriptions.

Features

• Context continuity: share the same project files, prompts, and metadata with every sub-agent
• Parallel execution: dispatch tasks concurrently to multiple model CLIs
• Model specialization: use the best model for each task (analysis, reasoning, code generation)
• Zero markup & no extra cost: uses your existing CLI tools/APIs with no intermediated billing
• IDE integrations: designed to work with many environments (Claude Code, Cursor, VS Code, JetBrains, etc.)
• Lightweight, local MCP server: runs on localhost and orchestrates sub-agent CLIs
• Simple CLI for configuration and agent selection

Installation / Configuration

Install via pip and run the server locally:

# Install the package
pip install roundtable-ai

# Start the server with default available agents
roundtable-ai

# Check what model CLIs the server can use on this machine
roundtable-ai --check

Start with a specific subset of agents:

# Start server using only Codex and Claude sub-agents
roundtable-ai --agents codex,claude

Add Roundtable as an MCP provider in a client (example: Claude Code):

# Add Roundtable as a Claude MCP agent (example CLI)
claude mcp add roundtable-ai -- roundtable-ai --agents gemini,claude,codex,cursor

Configuration tips:

• Ensure each model CLI you want to use is installed and configured with its API keys.
• The server runs on localhost and exposes a standard MCP-compatible endpoint for IDE assistants to call.
• Use the –check command to detect installed model CLIs and their reachable versions.

Available Tools / Resources

Roundtable delegates to whatever model CLIs are installed locally. Commonly used sub-agents include:

The server itself is available on GitHub: https://github.com/askbudi/roundtable. Use the repository for source, issue reporting, and contribution guidelines.

How it works (brief)

1. Your IDE’s primary assistant sends a single prompt that includes delegation instructions (e.g., “Use Gemini to scan for performance hotspots; use Codex to prepare a patch”).
2. Roundtable packages the shared file/project context and forwards the request to each specified sub-agent CLI in parallel.
3. Sub-agents run their analyses or produce outputs using local tooling (read files, run shell commands if allowed).
4. Roundtable aggregates and synthesizes the responses into a single structured reply for the primary assistant.

Use Cases

• Frontend performance triage

  • Prompt: “Gemini, analyze React components for expensive re-renders; Codex, propose a code change to memoize heavy components; Claude, summarize trade-offs and testing steps.”
  • Outcome: simultaneous analysis, a concrete patch, and a short QA checklist.

• Backend bottleneck investigation

  • Prompt: “Codex, inspect the /orders endpoint for N+1 queries and patch the ORM usage; Gemini, scan the repository for expensive queries; Claude, review infra logs for memory/CPU spikes.”
  • Outcome: targeted code fixes plus evidence from logs and repo scan.

• Pull request synthesis and testing

  • Prompt: “Claude, summarize the PR and list edge cases; Codex, generate unit tests; Cursor, apply suggested edits in files.”
  • Outcome: automated test generation and an edited branch ready for review.

• Cross-model research tasks

  • Use Gemini for large-context static analysis, Claude for design rationale, and a code model to implement results — all orchestrated from one IDE prompt.

Example prompt structure

Use simple delegation instructions inside your IDE assistant prompt:

The payment service intermittently times out.

- Use Gemini to scan backend services for long-running database calls.
- Use Codex to suggest and implement retry/backoff logic where appropriate.
- Use Claude to produce a rollout plan and monitoring metrics to watch.

Roundtable handles distributing this prompt, preserving file context, and returning combined outputs.

Contributing & License

Roundtable is open source (MIT). See the GitHub repo for contribution guidelines, issue tracking, and roadmap.