Agentset MCP
Retrieve from your Agentset RAG knowledge base directly inside Claude, Cursor, and other MCP clients.
Add to your client
Copy the config for your MCP client and paste it into its config file.
npx @agentset/mcpPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"agentset-mcp": {
"command": "npx",
"args": [
"-y",
"@agentset/mcp@latest"
],
"env": {
"AGENTSET_API_KEY": "agentset_xxx",
"AGENTSET_NAMESPACE_ID": "ns_xxx"
}
}
}
}Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- Node.js >= 18.17.0
- An Agentset account and API key (AGENTSET_API_KEY)
- An Agentset namespace id (AGENTSET_NAMESPACE_ID or the --ns flag) with ingested documents
About Agentset MCP
Agentset MCP bridges MCP clients to an Agentset RAG namespace. It performs semantic retrieval over your ingested documents and returns the matching text chunks as context for the model, so assistants can answer questions grounded in your own data. The server is launched via npx, talks over stdio, authenticates with an Agentset API key, and targets a specific namespace (and optionally a tenant).
Tools & capabilities (1)
knowledge-base-retrieveSearch the Agentset knowledge base (namespace) for data matching a query. Accepts a `query` string, an optional `topK` (1-100, default 10) for the maximum number of results, and an optional `rerank` boolean (default true) to reorder results by relevance. Returns the matching document text chunks. The tool description can be customized via the -d flag.
When to use it
- Let Claude or Cursor answer questions grounded in your private documents via RAG
- Add semantic search over an ingested document corpus to an AI assistant
- Build document-based, retrieval-augmented agent workflows without writing retrieval code
- Serve per-tenant knowledge in multi-tenant applications using the tenant id option
Security notes
Requires an Agentset API key (AGENTSET_API_KEY), which grants access to your Agentset account data. Provide it via the env block of your MCP client config and keep it out of source control. The server queries the namespace specified by AGENTSET_NAMESPACE_ID (or --ns); scope the API key and namespace appropriately, and use the tenant id (-t) to isolate data in multi-tenant deployments.
Agentset MCP FAQ
How do I tell the server which knowledge base to query?
Pass the namespace id with the --ns flag (e.g. `--ns your-namespace-id`) or set the AGENTSET_NAMESPACE_ID environment variable. The API key is provided via AGENTSET_API_KEY.
Can I use it in a multi-tenant setup?
Yes. Pass a tenant id with the -t/--tenant flag (e.g. `-t your-tenant-id`) so searches are scoped to that tenant's data.
Can I customize how the tool appears to the assistant?
Yes. Use the -d/--description flag to override the default tool description, e.g. `-d "Your custom tool description"`, which helps the model understand what knowledge it is querying.
What transport does it use?
It runs over stdio (StdioServerTransport), launched as a local process via npx, which is compatible with Claude Desktop, Cursor, Windsurf, and similar MCP clients.
Alternatives to Agentset MCP
Compare all alternatives →
Official MCP server providing persistent, file-backed knowledge-graph memory across sessions.
Structured step-by-step reasoning tool for breaking problems into revisable thought sequences.
Fully managed remote server for AWS docs, blogs, What's-New and Well-Architected guidance — no key.
Compare Agentset MCP with: