Agentic RAG for local and self-hosted document search: hybrid retrieval, reranking and multimodal RAG on embedded LanceDB, with Docling parsing and an MCP server
Setup
Choose your client, copy the config, then fill in any required environment variables.
claude mcp add haiku-rag -- uvx haiku-rag~/.codex/config.toml
[mcp_servers.haiku-rag]
command = "uvx"
args = ["haiku-rag"]
# codex mcp add haiku-rag -- uvx haiku-rag~/.cursor/mcp.json
{
"mcpServers": {
"haiku-rag": {
"command": "uvx",
"args": [
"haiku-rag"
]
}
}
}.vscode/mcp.json
{
"servers": {
"haiku-rag": {
"type": "stdio",
"command": "uvx",
"args": [
"haiku-rag"
]
}
}
}~/.gemini/settings.json
{
"mcpServers": {
"haiku-rag": {
"command": "uvx",
"args": [
"haiku-rag"
]
}
}
}claude_desktop_config.json
{
"mcpServers": {
"haiku-rag": {
"command": "uvx",
"args": [
"haiku-rag"
]
}
}
}.trae/mcp.json
{
"mcpServers": {
"haiku-rag": {
"command": "uvx",
"args": [
"haiku-rag"
]
}
}
}