generated: '2026-09-05' method: searched source: >- https://github.com/cloudera — four first-party MCP servers published under the Cloudera GitHub organization, each Apache-2.0 licensed, plus Cloudera's own MCP server framework (minimcp). deployment: mode: local-stdio install: uvx --from git+https://github.com/cloudera/cdv-mcp-server@main run-server package: https://github.com/cloudera/CDV-MCP-Server auth: api-key verified: searched note: >- THERE IS NO HOSTED CLOUDERA MCP ENDPOINT. Every server Cloudera publishes is a stdio process the operator installs and points at THEIR OWN Cloudera deployment — a CDV instance, a NiFi instance behind Knox, an Impala coordinator, or a Cloudera AI workbench. An agent cannot reach any of them without a human first installing the package and supplying a host plus credentials, so `mode` is local-stdio, not remote, and `endpoint` is deliberately empty. The `install` line above is the CDV server because it is the one that covers a whole product API; the three others are listed in servers[] with their own verbatim install lines. Note also that NONE of these servers front the CDP control plane API in openapi/ — the surface Cloudera ships as MCP and the surface it ships as Swagger are disjoint, which is what the tool crosswalk records. server_count: 4 servers: - name: Cloudera Data Visualization MCP Server slug: cdv-mcp-server repo: https://github.com/cloudera/CDV-MCP-Server license: Apache-2.0 language: python mode: local-stdio install: uvx --from git+https://github.com/cloudera/cdv-mcp-server@main run-server transports: [stdio, http, sse] transport_note: >- Selected with the MCP_TRANSPORT environment variable; `http` and `sse` are the operator's own self-hosted process, not a Cloudera-hosted endpoint. auth: api-key config: - CDV_BASE_URL tool_count: 52 covers: >- The full Cloudera Data Visualization REST API — groups, users, roles, segments, filter associations, workspaces, datasets, visuals, data connections, jobs, extracts, the data API, and operational debugging. tools: [get_group, create_group, update_group, delete_group, get_user, update_user, edit_user_profile, get_role, create_role, update_role, delete_role, get_segment, create_segment, update_segment, delete_segment, get_filter_association, create_filter_association, update_filter_association, delete_filter_association, get_workspace, create_workspace, update_workspace, delete_workspace, get_dataset, create_dataset, update_dataset, delete_dataset, get_visual, create_visual, update_visual, delete_visual, create_smart_visual, get_connection, create_connection, update_connection, delete_connection, export_connection, query_dataapi_get, query_dataapi_post, query_data_get, query_data_post, run_job, run_extract, create_extract, post_gc_monitor, post_gc_stats, set_log_level, get_logger_level, set_logger_level, toggle_cprofile, reset_dataconnection_cache, reset_dataset_cache] tools_source: >- Read from the tool tables in the repository README; the live inputSchema for each tool requires running the server against a CDV instance. - name: Cloudera AI Workbench MCP Server slug: cai-workbench-mcp-server repo: https://github.com/cloudera/CAI_Workbench_MCP_Server license: Apache-2.0 language: python mode: local-stdio install: >- uvx --from git+https://github.com/cloudera/CAI_Workbench_MCP_Server.git --with https://ml-xxxx.cloudera.site/api/v2/python.tar.gz cai-workbench-mcp-stdio transports: [stdio, http] auth: api-key config: - CAI_WORKBENCH_HOST - CAI_WORKBENCH_API_KEY - CAI_WORKBENCH_PROJECT_ID tool_count: 106 covers: >- Cloudera AI (CML) workbench — project files, jobs and job runs, model build and deployment, experiment tracking, applications, runtimes, quotas and collaborators. Every tool is a thin wrapper over the official cmlapi Python SDK (CMLServiceApi) rather than raw HTTP. tools_source: >- 106 *_tool names enumerated from the repository README. Cloudera's HTTP transport mode is documented as having no authentication and is described for development/testing only. - name: NiFi MCP Server (via Knox) slug: nifi-mcp-server repo: https://github.com/cloudera/NiFi-MCP-Server license: Apache-2.0 language: python mode: local-stdio install: uvx --from git+https://github.com/cloudera/nifi-mcp-server@main transports: [stdio] auth: oauth auth_note: >- Apache Knox bearer tokens, cookies or passcode tokens for CDP deployments. config: - NIFI_API_BASE - NIFI_READONLY tool_count: 66 tool_breakdown: read_only: 24 write: 42 covers: >- Apache NiFi 1.x and 2.x flow inspection and construction — process groups, processors, connections, controller services, ports, parameter contexts, bulletins and flow health. Read-only by default; the 42 write operations require NIFI_READONLY=false. safety_note: >- Cloudera's own README flags empty_connection_queue as data loss and terminate_processor as a last resort. This is the strongest agent-safety signal in Cloudera's MCP surface and it is prose in a README, not a machine-readable consequence annotation. - name: Cloudera Iceberg MCP Server (via Impala) slug: iceberg-mcp-server repo: https://github.com/cloudera/iceberg-mcp-server license: Apache-2.0 language: python mode: local-stdio install: uvx --from git+https://github.com/cloudera/iceberg-mcp-server@main run-server transports: [stdio] auth: api-key auth_note: IMPALA_USER / IMPALA_PASSWORD against an Impala coordinator. config: - IMPALA_HOST - IMPALA_PORT - IMPALA_USER - IMPALA_PASSWORD - IMPALA_DATABASE tool_count: 2 read_only: true covers: Read-only access to Iceberg tables through Apache Impala. tools: - name: execute_query description: Run any SQL query on Impala and return the results as JSON. - name: get_schema description: List all tables available in the current database. related: - name: minimcp repo: https://github.com/cloudera/minimcp package: https://pypi.org/project/minimcp/ version: 0.5.2 published: '2026-03-23' kind: framework note: >- Cloudera also publishes its own minimal, stateless MCP server framework. It is not itself an MCP server for a Cloudera product, so it is recorded here rather than in servers[]. gaps: - >- No MCP server exists for the CDP Public Cloud control plane (environments, datalake, datahub, dw, de, df, iam) — the 778-operation Swagger surface in openapi/ has no agent binding at all. - >- No tool inputSchema could be captured: every server requires a customer deployment plus credentials before tools/list will answer, so tool names and descriptions here come from Cloudera's own READMEs and the schemas would need authenticated introspection against a live instance.