MCP server
Use AI assistants to search through Exacta indexes.
What this is
Exacta exposes its search engine over the Model Context Protocol (MCP). Point any MCP-compatible AI client at the endpoint below and the model can list indexes, run searches, and pull documents on your behalf. Free, no account, no API key, no rate-limit ceiling beyond what the public API already enforces.
Endpoint
https://mcp.polymood.netTransport: streamable HTTP. Read-only. No authentication.
Tools
list_indexes()Discover available document archives with descriptions, tags, and document counts.
get_index_schema(index)Get the field schema for an index (which fields exist, their types, low-cardinality values).
search(index, query, search_fields?, exact?, fuzzy?, fuzzy_distance?, page?, limit?, sort_desc?)Run a search. Supports default ranked relevance, exact phrase mode, and fuzzy/typo-tolerant mode. Returns snippets with <mark> highlighting and document addresses.
get_document(index, segment_ord, doc)Fetch the full content of a single document by the address returned in search results.
list_field_values(index, field, prefix?, limit?)Enumerate distinct values of a text field. Useful for building filters or exploring the vocabulary of a field.
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"exacta": {
"url": "https://mcp.polymood.net",
"transport": "streamable-http"
}
}
}Cursor
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"exacta": {
"url": "https://mcp.polymood.net"
}
}
}Other clients
Any client that speaks streamable HTTP MCP can connect. Codex CLI, Continue, Cline, and others ship with MCP support. Point them at the endpoint above.
Example session
Once connected, you can ask the model things like:
- “What indexes are available on Exacta?”
- “Search WorldRadioHistory for articles about cavity resonators in 1962.”
- “Find Spex reviews of Test Department from the 1980s and quote the snippets.”
- “Show the full body of the top result.”
Limits and fair use
The MCP server forwards requests to the public Exacta API. The same per-IP rate limits apply. If you plan to run something at scale please get in touch first.