> ## Documentation Index
> Fetch the complete documentation index at: https://webscraping.titannet.io/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Titan MCP server

> Give AI agents live web access through Titan: search the open web, fetch clean page content, and crawl sites, using the same credits and API keys as the rest of the platform.

The **Titan MCP server** exposes Titan's web intelligence platform as [Model Context Protocol](https://modelcontextprotocol.io) tools. Point an MCP-compatible client at it and your agent can search the web, read pages as clean markdown, and map sites—without you writing an integration or managing scrapers.

It is a hosted remote server. There is nothing to install and nothing to run locally.

```text theme={null}
https://mcp.webscraping.titannet.io/mcp
```

<Note>
  New here? Follow the [Quickstart](/docs/mcp/quickstart) to connect a client and make your first tool call in a few minutes.
</Note>

## What your agent can do

<CardGroup cols={2}>
  <Card title="Search the web" icon="magnifying-glass" href="/docs/mcp/tools/titan-search">
    Query major search platforms with real search operators and get ranked organic results.
  </Card>

  <Card title="Read pages" icon="file-lines" href="/docs/mcp/tools/titan-fetch">
    Turn up to 100 URLs into clean markdown or text, with boilerplate stripped out.
  </Card>

  <Card title="Map and crawl sites" icon="sitemap" href="/docs/mcp/tools/titan-crawl">
    Discover a site's URL inventory, or crawl it with content extraction, inside explicit limits.
  </Card>

  <Card title="Run curated templates" icon="layer-group" href="/docs/mcp/tools/titan-run-template">
    Execute Titan's allowlisted extraction templates when a generic fetch is not specific enough.
  </Card>
</CardGroup>

## Why route agents through Titan

An agent that fetches pages directly hits the problems every scraping team hits: blocks, JavaScript-rendered pages, rate limits, inconsistent HTML, and no record of what was retrieved. Titan handles those on the way through.

| What you get              | Why it matters to an agent                                                                                       |
| ------------------------- | ---------------------------------------------------------------------------------------------------------------- |
| **Distributed execution** | Requests run across Titan's worker network, not from your agent's IP.                                            |
| **Real browsers**         | JavaScript-rendered pages resolve before extraction, so single-page apps return content instead of empty shells. |
| **Normalized output**     | Every result is typed and consistent across providers, so prompts do not need per-site parsing.                  |
| **Bounded cost**          | Every call reports estimated and consumed credits, and hard caps stop runaway crawls.                            |
| **Full audit trail**      | Every tool call becomes a Titan execution you can inspect, export, and reconcile against billing.                |

## The tool surface

Six tools, deliberately small. Agents choose well when the choice is narrow.

| Tool                                                      | Use it to                                |
| --------------------------------------------------------- | ---------------------------------------- |
| [`titan_search`](/docs/mcp/tools/titan-search)                 | Find URLs when you do not have them      |
| [`titan_fetch`](/docs/mcp/tools/titan-fetch)                   | Read content from URLs you already have  |
| [`titan_crawl`](/docs/mcp/tools/titan-crawl)                   | Discover or extract a site's pages       |
| [`titan_list_templates`](/docs/mcp/tools/titan-list-templates) | See which curated templates are runnable |
| [`titan_run_template`](/docs/mcp/tools/titan-run-template)     | Run a curated template by slug           |
| [`titan_get_run`](/docs/mcp/tools/titan-get-run)               | Poll a run and retrieve its results      |

The usual pattern is **search, then fetch**: discover candidate URLs, pick the promising ones, read only those. Crawling a whole site is the exception, not the default.

## How a tool call works

```mermaid theme={null}
sequenceDiagram
    participant Agent as Your agent
    participant MCP as Titan MCP
    participant Titan as Titan platform
    participant Web
    Agent->>MCP: titan_search("...")
    MCP->>Titan: Create run
    Titan->>Web: Execute on worker network
    Web-->>Titan: Raw pages
    Titan-->>MCP: Normalized records
    MCP-->>Agent: results[] + run_id + usage
```

Short runs return results in the same call. Longer runs return a `run_id` immediately and your agent calls [`titan_get_run`](/docs/mcp/tools/titan-get-run) when it is ready. See [Runs and results](/docs/mcp/runs-and-results).

## What it costs

Agent calls draw from the same credit wallet as the rest of Titan, and **you are charged for successful delivery only**. A search that gets blocked, a URL that fails to load, a crawl that returns nothing—none of those consume credits.

See [Credits and usage](/docs/mcp/credits-and-usage) for the per-tool rules.

## Relationship to the rest of Titan

The MCP server is a front door, not a separate product. Behind every tool call is an ordinary Titan task and execution:

* The MCP `run_id` **is** the Titan `execution_id`. Use it in the [Task Service API](/docs/use-the-platform/use-the-task-service-api), analytics, and billing.
* Agent runs appear in your task list, filterable as [agentic tasks](/docs/about-platform/tasks#agentic-tasks).
* Results, datasets, and exports work exactly as they do for tasks you create yourself.

That means you can prototype with an agent and productionize with the API, against the same objects.

## Get started

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/docs/mcp/quickstart">
    Create a key, connect a client, run your first search.
  </Card>

  <Card title="Connect your client" icon="plug" href="/docs/mcp/connect-your-client">
    Setup for Claude Code, Claude Desktop, Cursor, VS Code, and Windsurf.
  </Card>

  <Card title="Authentication and scopes" icon="key" href="/docs/mcp/authentication-and-scopes">
    Which `mcp:*` scopes each tool needs, and how to scope a key tightly.
  </Card>

  <Card title="Tool reference" icon="wrench" href="/docs/mcp/tools/overview">
    Full parameters, defaults, limits, and response shapes.
  </Card>
</CardGroup>