data-ai MCP Server
Zero-configuration MCP server that unifies multiple AI coding assistants (Codex, Claude Code, Cursor, Gemini) through intelligent auto-discovery and standardized interface
Discovered via unknown and last synced 3mo ago.
1. Install the package
uvx roundtable-ai@latest
2. Add to claude_desktop_config.json
{
"mcpServers": {
"roundtable": {
"command": "npx",
"args": [
"roundtable"
]
}
}
}Config file location: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) / %APPDATA%\Claude\claude_desktop_config.json (Windows)
Roundtable is ideal for engineering teams and individual developers who use multiple AI coding assistants (Claude Code, Cursor, Gemini, Codex) and want a unified interface without complex configuration. DevOps engineers, full-stack developers, and AI-first teams benefit most by reducing context-switching and standardizing their AI-assisted workflows across different tools.
Access Codex, Claude Code, Cursor, and Gemini through a single standardized protocol instead of managing separate integrations. This reduces cognitive load when switching between tools during development sessions.
Roundtable auto-discovers available AI coding assistants in your environment and automatically configures them. No manual setup required—start using unified AI capabilities immediately after server initialization.
Establish consistent interfaces and interaction patterns for AI-assisted development across your entire engineering team, regardless of which underlying assistant each developer prefers.
Build workflows and automations against a standardized interface rather than specific AI assistants, making it easier to migrate between tools or use multiple assistants simultaneously without rewriting integrations.
No. Roundtable uses intelligent auto-discovery to detect available assistants in your environment and automatically standardizes their interfaces. Zero configuration is required to get started.
Roundtable currently supports Claude Code, Cursor, Gemini, and Codex. The auto-discovery system means you only pay attention to the assistants you actually have installed or available in your environment.
Yes. Roundtable acts as an MCP server that integrates with Claude Code and other compatible clients. Your existing workflows continue to work while gaining access to other AI assistants through the unified interface.
Roundtable adds minimal overhead since it's a standardized protocol layer. The main benefit is eliminating context-switching between different interfaces, which typically saves more time than any performance cost.
`false`
Description
Databricks personal access token
8
`Roundtable`
Server-level OpenAI key
`default`
Vertex AI region (default: `us-central1`)
Default Snowflake database
SQL safety for BigQuery, Snowflake, Databricks
Database
`false`
SQL warehouse HTTP path
Fetch, HTML stripping, truncation
dev default
Server-level Google AI key
Snowflake password
Tests
(none)
BigQuery — uses same ADC as Vertex AI
Snowflake compute warehouse
Allowlist, dangerous patterns, path traversal
Use Case
BigQuery — max bytes scanned per query (default: 1GB)
Default Unity Catalog
Session validation
Snowflake username
Description
Export validation
Databricks workspace URL
Read, write, list, find + path traversal
Description
9
Server-level Anthropic key
Google Custom Search API key (for web_search tool)
BigQuery dataset location (default: `US`)
Snowflake account identifier
JSON-RPC 2.0 compliance, initialize/tools/list/call, error codes
Build, verify, tamper detection, hash matching
`false`
Override BigQuery billing project (cross-project access, default: `GCP_PROJECT` value)
SQL fusion, dedup, LIMIT injection
`false`
Google Custom Search engine ID
Canonicalize, build, verify, decrypt, sign, expiry
Query, tool_call, discover, auth gate, SQL safety
HKDF key derivation, AES-256-GCM encrypt/decrypt, HMAC signing, timestamp freshness
`false`
Default Ollama host URL (default: `http://localhost:11434`, overridable per-workspace)
Replay detection, TTL expiry, cleanup
All 22+ tools validated, resolveTools filtering, OpenAI/Anthropic/Google format output
`3000`
HMAC auth validation, contract enforcement, expired timestamps