PuriliMCP
LIVESTREAMABLE HTTPREAD ONLY

Free web search for any MCP client.

Give an AI application access to ranked web results and stored page context from Purili's independent index. One hosted endpoint, six tools, no account, and no API key.

Connect a client Protocol documentation
No API keyNo fixed quotaIndependent indexCORS-free server connection
PURILI MCPONLINE

endpoint https://puri.li/mcp

transport streamable-http

auth none

tools/list

✓web_search

✓search_domain

✓get_context

✓suggest_queries

✓get_infocard

✓get_index_stats

Connect

Use the hosted endpoint.

Add Purili as a remote Streamable HTTP server in a compatible MCP client. The exact settings screen differs by client; the server URL never changes.

Remote MCP server
https://puri.li/mcp
Generic remote configuration
{
  "mcpServers": {
    "purili": {
      "url": "https://puri.li/mcp"
    }
  }
}

This URL is for an MCP client

It is not a REST endpoint or human webpage. MCP clients negotiate the protocol, discover the tools, validate their arguments, and call them when needed.

Available tools

Search, select, then read.

The toolset is intentionally small. Models get clear choices instead of dozens of overlapping actions.

web_search

Search the independent Purili web index and return ranked sources.

search_domain

Restrict a search to one domain and its subdomains.

get_context

Retrieve stored page text, passages, metadata, and coverage.

suggest_queries

Get clean query completions for a partial search.

get_infocard

Retrieve a confident entity card or instant answer.

get_index_stats

Read public crawler and index statistics.

Our own index

Core web results come from Purili's crawler and independent index, not a resold search feed.

Nothing to provision

No signup, API-key exchange, secrets manager, billing account, or per-client configuration.

Context with provenance

Stored text includes its source URL, coverage label, index provenance, and liveFetch: false.

How it works

MCP is the adapter.

  1. 1. Your client connects

    The MCP client opens https://puri.li/mcp and discovers the available tool schemas.

  2. 2. The model chooses a tool

    For a research question it can search, restrict by domain, or request stored context for a result.

  3. 3. Purili queries its services

    The server calls the same public APIs documented for developers and returns structured MCP content.

  4. 4. The client receives evidence

    The model gets ranked URLs or stored passages it can use as explicitly sourced, untrusted evidence.

FAQ

About the Purili MCP server

What is the Purili MCP server?+

It is a Model Context Protocol server that exposes Purili web search and indexed page context as tools an AI application can call.

Why did /mcp show a JSON error in my browser?+

The /mcp URL is a machine endpoint that expects MCP protocol requests, not an ordinary webpage. Browser visits now redirect to this guide while MCP clients continue using /mcp directly.

Do I need an API key?+

No. The hosted Purili MCP server is public and does not require a Purili account, token, or API key.

Does Purili use Google, Bing, or Brave for core web search?+

No. The core Search tools use Purili's own crawler and independent web index.

Does get_context scrape a page live?+

No. It returns content already stored in Purili's index and explicitly reports liveFetch as false.

Is the MCP server read-only?+

Yes. Every currently exposed tool reads public search or index information and makes no external changes.

Can I install it through npm?+

Yes. Install @purili/mcp-server from npm or run it directly with npx. It exposes the same six tools as the hosted endpoint.

Connect Purili to your MCP client.

https://puri.li/mcp · no authentication required

Connection details