MCP Server

The WDS API Server exposes a Model Context Protocol endpoint for IDEs and agentic systems. Its current surface consists of 21 registered tools for job, download, Traversal, scheduling, tenant, and retrieval workflows, plus seven registered prompts that compose those tools into common tasks. See Plans for feature availability.

Connect to the MCP Server

Deploy the WDS API Server, then connect an MCP client to its HTTP endpoint.

A deployment mounted below a base path must prepend that path to /mcp. For example, a base path of /wds produces /wds/mcp.

For Visual Studio Code setup, see its MCP server documentation.

The server identifies itself as WDS MCP Server and reports the application assembly version. Clients must support stateless Streamable HTTP; do not depend on a server-maintained session or an Mcp-Session-Id for later calls.

Tools

The MCP Tools catalog documents all registered names, arguments, defaults, return schemas, and behavior flags.

Current capabilities include:

• Manage jobs with WdsGetJobs, WdsGetJobConfig, WdsUpsertJobConfig, and WdsDeleteJob.
• Fetch one page with WdsFetchChunked, then poll and read its content with WdsGetChunk.
• Store and run hierarchical plans with WdsUpsertTraversalConfig, WdsRunTraversal, and the Traversal inspection tools.
• Configure recurring runs with the three Wds*TraversalSchedule tools when Scheduling is enabled.
• Search retrieval-indexed content with WdsRetrieve when Retrieval is enabled.
• Remove all data for the default MCP tenant with WdsDeleteTenant.

Prompts

The MCP Prompts catalog documents the seven exact registered prompt names:

• assess proposes a Traversal extraction plan; scrape runs an already configured plan and returns its data.
• resume and sliced-resume crawl and summarize a site.
• index and reindex enroll content for retrieval; query answers a task from that index.

Typical Flows

Fetch One Page

1. Persist a job with WdsUpsertJobConfig.
2. Call WdsFetchChunked with the job name and absolute URL.
3. Poll WdsGetChunk with the returned task ID until content is available.
4. Continue reading spans with startIndex and length when the page is larger than one response.

Run a Traversal

1. Persist JobConfig with WdsUpsertJobConfig.
2. Persist TraversalConfig with WdsUpsertTraversalConfig.
3. Start the run with WdsRunTraversal and retain its runNum.
4. Poll WdsGetTraversalRunInfo, inspect WdsGetTraversalRunErrors, and page through WdsGetTraversalRunData until dataCursor is null.

Set root-level TraversalConfig.MaxRuns through WdsUpsertTraversalConfig to limit completed-run history. Starting a new run can remove older completed runs, their results, and unshared task data; save needed results before continuing. Branch configs and returned run-plan snapshots do not contain MaxRuns. This limits run history rather than total storage; see retention limitations.

Use Retrieval Prompts

1. Invoke index, or reindex when cached content must be reenrolled.
2. After enrollment completes, invoke query with the target URL and task.