# MCP Server | Parsium Docs

> Connect agent clients to the Parseium runtime over MCP.

Source: https://docs.parseium.com/guides/mcp/

# MCP Server

Parseium exposes Streamable HTTP MCP at:

```
POST https://api.parseium.com/mcp
```

Authenticate with `Authorization: Bearer <api-key>` or `X-API-Key`. Deployments can optionally enable Clerk OAuth; discover protected-resource metadata at `GET /.well-known/oauth-protected-resource/mcp`. OAuth never enables REST access—`/v1/*` remains API-key only.

## Tools

 Section titled “Tools”

 Tool Purpose list_parsers , get_parser Discover owned parsers create_parser , update_parser , delete_parser Manage parser definitions test_parser Dry-run inline or stored parser configs scrape Execute one parser scrape execute_prebuilt Call a prebuilt API create_batch Submit parser or prebuilt batch work get_job Read j_ , b_ , or wr_ status get_status Read credits, concurrency, pending work, and active batches list_databases , create_database , query_database Manage/query datasets manage_cron List/create/update/delete/run cron resources manage_workflow , test_workflow Manage, run, and preflight workflows

## Result limits

 Section titled “Result limits”

MCP responses are designed for agent context windows. Large scrape, prebuilt, and SQL outputs are truncated. Use HTTP batches with database sinks and bounded SQL queries for bulk data. Table dump has no MCP tool by design.

`create_batch` can read UTF-8 lines or CSV from `file_path` when the server administrator has allowed that location through `MCP_BATCH_FILE_ROOTS`.

MCP tools enforce the same ownership, billing, validation, and queue limits as their HTTP counterparts.