MCP server
Convex MCP server
Read tables, run functions and inspect logs of a Convex deployment.
★ 13kpushed 2026-10-24docson DevHunt
Run in a terminal. Everything after -- is the command that starts the server. docs
claude mcp add --transport stdio convex -- npx -y convex@latest mcp start
Claude Desktop only: Settings > Developer > Edit Config opens claude_desktop_config.json. Add this, save and restart the app. docs
{
"mcpServers": {
"convex": {
"command": "npx",
"args": [
"-y",
"convex@latest",
"mcp",
"start"
]
}
}
}Add to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (this project), or use the install button. docs
{
"mcpServers": {
"convex": {
"command": "npx",
"args": [
"-y",
"convex@latest",
"mcp",
"start"
]
}
}
}Add to .vscode/mcp.json (note the "servers" key and the type), or use the install button. Command Palette > "MCP: Open User Configuration" for all workspaces. docs
{
"servers": {
"convex": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"convex@latest",
"mcp",
"start"
]
}
}
}Run codex mcp add convex -- npx -y convex@latest mcp start, or add this to ~/.codex/config.toml. docs
[mcp_servers.convex] command = "npx" args = ["-y", "convex@latest", "mcp", "start"]
Run in a terminal (-s user adds it for all projects; the default is this project). docs
gemini mcp add -s user convex npx -- -y convex@latest mcp start
Run in a terminal; it writes ~/.copilot/mcp-config.json. docs
copilot mcp add convex -- npx -y convex@latest mcp start
Add to mcp_config.json: ~/.config/devin/mcp_config.json on macOS and Linux (older installs: ~/.codeium/windsurf/mcp_config.json), %APPDATA%\devin\ on Windows. docs
{
"mcpServers": {
"convex": {
"command": "npx",
"args": [
"-y",
"convex@latest",
"mcp",
"start"
]
}
}
}Add to Zed's settings.json (or Settings > AI > MCP Servers > Add Server). Remote servers without an Authorization header sign in with OAuth. docs
{
"context_servers": {
"convex": {
"command": "npx",
"args": [
"-y",
"convex@latest",
"mcp",
"start"
]
}
}
}Add to opencode.json in the project (or ~/.config/opencode/opencode.json). docs
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"convex": {
"type": "local",
"command": [
"npx",
"-y",
"convex@latest",
"mcp",
"start"
],
"enabled": true
}
}
}Add to ~/.kiro/settings/mcp.json (all projects) or .kiro/settings/mcp.json (this project). docs
{
"mcpServers": {
"convex": {
"command": "npx",
"args": [
"-y",
"convex@latest",
"mcp",
"start"
]
}
}
}MCP Servers > Configure > Configure MCP Servers opens cline_mcp_settings.json; add this entry. docs
{
"mcpServers": {
"convex": {
"command": "npx",
"args": [
"-y",
"convex@latest",
"mcp",
"start"
]
}
}
}Agent skills for Convex
The MCP server gives the agent Convex's tools; a skill teaches it how to use Convex well. Install with the skills CLI or see each skill's page for Claude Code, Codex, Copilot and Cursor.
- ConvexConvex
Guidance for building on Convex's TypeScript backend: database, functions, scheduling, sync.
npx skills add convex/agent-skills --skill convex
FAQ
Is there an official Convex MCP server?+
Yes: Convex, maintained by Convex.
How do I connect Convex to Claude?+
In Claude Code, run the claude mcp add command shown above. In Claude Desktop, add the server to claude_desktop_config.json (Settings > Developer > Edit Config) and restart the app.
How do I use the Convex MCP server in Cursor or VS Code?+
Open the Cursor or VS Code tab on the server above: it has the exact mcp.json entry and a one-click install link. The same server also works in Claude Code, Claude Desktop / claude.ai, ChatGPT, Cursor, VS Code and Codex.
Is it safe to give an AI agent access to Convex?+
Start with the official server, a read-only or test account and approval prompts on for anything that writes. Content the agent reads from Convex (tickets, messages, pages) can carry prompt-injection instructions, so don't auto-approve actions.
All database MCP servers · Agent skills · How MCP setup works in each client
MCP servers for similar apps: AlloyDB · Azure SQL · BigQuery · Chroma · ClickHouse · Cloud SQL · CockroachDB · Confluent · Couchbase · Databricks · Dataplex · dbt · DuckDB · Elasticsearch · Firebase · Firestore · Fivetran · Kafka · MariaDB · Metabase · Milvus · MongoDB · MotherDuck · MySQL