MCP server
Give agents a direct path to structured memory.
The RushDB MCP server exposes database operations as tools for MCP-compatible clients. Agents can discover structure before they build queries.
Agent integrations drift when every client gets custom glue code.
A handwritten tool layer needs to expose records, queries, relationships, and error handling consistently. Every new client becomes another integration to maintain.
Before
- Build custom database tools for each agent
- Teach query syntax in long prompt files
- Reimplement schema discovery
- Maintain separate integration code per client
With RushDB
- Connect an MCP-compatible client
- Browse labels and properties through tools
- Query and mutate records through one server
- Use discovery-first guidance before complex queries
What it enables
Give agents tools without rebuilding the backend.
The MCP server is the fastest way to expose RushDB to assistants and coding agents. The important part is not just access; it is discovery-first access that reduces hallucinated labels, fields, and query operators.
Hosted or local connection
Connect ChatGPT, Claude.ai, and other web assistants to the hosted OAuth endpoint at mcp.rushdb.com/mcp — one URL, sign in, no install. Desktop and coding clients run the local npx server.
Database operations as tools
Agents can browse structure, query records, manage relationships, run bulk operations, export data, and use transactions.
Discovery before execution
The documented workflow starts with schema, classifies intent, loads the query spec when needed, and then builds the tool call.
How it works
Start with the smallest useful path.
01
Add the MCP server
Configure the local package or a hosted MCP connection in a compatible client.
02
Discover before querying
Use labels, properties, schema, or the query-builder prompt so the agent works against the real project shape.
03
Keep operations structured
Create records, attach relationships, run searches, export data, and manage transactions through explicit tools.
Flow
Agent request to grounded query
RushDB MCP is designed around discovery before action, so the agent checks available structure before constructing record searches or mutations.
Implementation sketch
Connect hosted with OAuth, or run the local server.
The hosted endpoint needs no install: add the URL as a connector, sign in with your RushDB account, and pick a project. Desktop and coding clients run the local server below; exact config placement varies by MCP client.
Hosted MCP with OAuth
https://mcp.rushdb.com/mcp
Add this URL as a connector in ChatGPT, Claude.ai, or any web client, then sign in. No install, no API key handling.
# Local MCP server (desktop and coding clients): RUSHDB_API_KEY=your-key npx -y @rushdb/mcp-server
Know the operational boundary.
MCP-compatible clients
Compatibility depends on the client implementing MCP. Use the setup path documented for the client you are connecting.
Read the MCP server guideSkills complement runtime tools
MCP gives agents tools. Agent skills can add task-specific instructions for querying, modeling, and persistent memory workflows.
Read about agent skillsNext step
Build one focused workflow.
Related features
Explore all featuresSchema API
Expose live labels, property types, sample values, numeric ranges, relationship directions, and vector-index status to agents and apps.
See schemaUnified query API
Use one SearchQuery-shaped contract for records, labels, relationships, property metadata, and distinct values or ranges.
See SearchQueryACID transactions
Group related reads and writes into a single atomic unit. Commit the complete change or roll it back without partial state.
See safe writes