Hybrid Manager exposes a Model Context Protocol (MCP) server that allows AI agents and MCP-compatible clients to interact with the platform programmatically. The MCP server enables agentic workflows (such as automated migrations, cluster management, and observability queries) to be driven by large language models (LLMs) or AI pipelines.
The MCP server is self-documenting. When an MCP client connects to the entry point, it can automatically discover all available tools and their descriptions without requiring separate API reference documentation.
Connecting to the MCP server
The MCP server is available via the Portal API Gateway. Use the following entry point URL, replacing <portal-url> with your HM environment's portal URL:
https://<portal-url>/api/mcp
Requests to the MCP server must include the following headers:
| Header | Value |
|---|---|
Authorization | Bearer <access-key> |
Accept | application/json, text/event-stream |
Use your portal access key. You must have sufficient permissions. For information on how to create an access key, see Access key.
Available tool categories
The MCP server exposes tools across these functional areas:
- Cluster management — Create, view, scale, update, and delete Postgres clusters (PG, PGE, and EPAS) in Single Node, High Availability, Advanced High Availability, and Distributed High Availability topologies, manage cluster templates, backups, and maintenance windows.
- Observability and monitoring — Retrieve health scores, current and historical metrics, active alerts, logs, and Grafana dashboards.
- Recommendations — Get index, statistics, configuration tuning, and security recommendations for clusters.
- Migrations — Manage database migrations end-to-end, including assessment, schema conversion, Oracle SQL translation, and migration lifecycle (create, pause, resume, complete).
- Lakehouse and analytics — Manage catalogs, tiered tables, and data offload for lakehouse workloads.
- Agent Factory — Configure inference services, models, model server clusters, knowledge bases, pipelines, and Langflow-based AI flows.
- Identity and access — Manage access keys, machine users, identity providers (IdPs), and organization and project roles.
- Projects and organization — Create and manage projects, tags, locations, and view usage and activity logs.
- Database introspection — Browse schemas, tables, partitions, DDL, and analyze query performance directly through the MCP interface.
Accessing Postgres cluster databases via MCP
The HM MCP server manages Hybrid Manager operations. It does not provide direct MCP access to the Postgres databases running inside the clusters it manages. To interact with those databases using MCP — for example, to run queries, tune indexes, or analyze query plans — use Pg Airman MCP.
Tool limits with LLM clients
Most LLMs enforce a hard limit of 100–200 tools per session. The HM MCP server exposes a large number of tools. When using an LLM-based MCP client, you may need to select a subset of tools relevant to your use case and disable others. Refer to your MCP client's documentation for instructions on how to enable or disable individual tools.