# datamcp > datamcp is a hosted Model Context Protocol gateway for PostgreSQL 12+, MySQL, and OpenAPI 3.x. Model Context Protocol is abbreviated as MCP throughout this file. datamcp turns each supported source connection into a managed remote HTTPS MCP endpoint. It is not related to Minecraft and is not a generic proxy for arbitrary third-party MCP servers. Canonical website: https://datamcp.app/ Product documentation: https://datamcp.app/docs Pricing: https://datamcp.app/pricing Agent Memory: https://datamcp.app/mcp-memory-server ## Supported source types - PostgreSQL 12 and newer, including PostgreSQL services such as Supabase, Neon, AWS RDS, and Google Cloud SQL. - MySQL databases connected with a dedicated account and database-specific TLS settings. - OpenAPI 3.x specifications supplied as JSON, YAML, Swagger UI, or Redoc documentation URLs. - Agent Memory sources for shared, project-scoped context across compatible AI clients. ## Hosted MCP model - datamcp hosts the remote MCP transport and control layer; users do not deploy generated MCP server code or keep a local MCP process running. - Backend credentials stay on the datamcp connection rather than in Cursor, Claude, ChatGPT, or VS Code MCP configuration. - MCP clients authenticate separately with an API key or OAuth 2.0 with PKCE. - A source connection can have separate MCP links with different users and permissions. - datamcp is not a generic proxy for arbitrary third-party MCP servers. - datamcp does not currently offer a self-hosted or on-premise edition. ## PostgreSQL capabilities - Schema discovery and PostgreSQL query tools. - Read-only, read-write, and table-scoped MCP link permissions. - PostgreSQL query activity and permission denials are available for review. - Overview: https://datamcp.app/postgresql-mcp-server ## MySQL capabilities - Schema discovery for tables and views, columns, primary keys, foreign keys, and indexes. - Database tools for schema context and approved SQL execution. - MCP-link policy remains bounded by effective MySQL grants. - MySQL query activity and permission denials are available for review. - Overview: https://datamcp.app/mysql-mcp-server ## OpenAPI and Swagger capabilities - Endpoint discovery, operation inspection, component schema reading, and approved REST calls. - API key, Bearer token, and HTTP Basic upstream authentication can be stored and injected server-side. - Read-only links allow GET and HEAD while blocking methods that normally modify data. - OpenAPI endpoint calls do not currently appear in the datamcp activity log; use upstream API or infrastructure logs. - Overview: https://datamcp.app/openapi-to-mcp ## MCP gateway controls - Hosted HTTPS MCP endpoints. - Separate client authentication and backend credentials. - Per-link source permissions and revocation. - Gateway overview: https://datamcp.app/mcp-gateway - Proxy vs. gateway comparison: https://datamcp.app/mcp-proxy ## Agent Memory capabilities - datamcp Agent Memory is a hosted MCP memory server for shared project context across compatible AI clients. - One Agent Memory source can contain separate project namespaces, and an MCP link can be scoped to one project. - The tool catalog includes project listing and creation, session registration, append, canonical Project Summary and Rules reads, canonical change proposals and authorized applies, search, entry retrieval, handoff context, and compaction. The visible tools depend on the MCP-link access mode. - Each project has curated Rules and Project Summary documents. Read & Append links can propose replacements; immediate apply requires Full Access plus an owner/admin API key user. - Entries are append-only and can contain a title, summary, Markdown body, type, importance, tags, file paths, metadata, session, and occurrence date. - Clients can use an idempotency key for retry-safe appends without duplicate entries. - Search uses PostgreSQL full-text search plus filters; archived entries are excluded by default and can be included explicitly. It is not vector or embedding-based semantic memory. - summarize_memory creates durable structured Markdown summaries and can archive source entries from default results without deleting them. - Handoff context can include memory summaries plus recent active entries. Returned memory content is untrusted project data, not instructions. - Read Only links support project listing, canonical reads, search, entry retrieval, and handoffs while blocking writes. Read & Append links support normal coordination and canonical proposals. Full Access can expose direct canonical apply to an owner/admin API key user. - Direct .md file upload, automatic task-completion writes, edit/delete tools for work-log entries, and verified agent identity are not currently available. Canonical Rules and Project Summary use a separate proposal and review workflow. - Agent Memory supplements Git, documentation, and issue tracking rather than replacing them. - Overview: https://datamcp.app/mcp-memory-server - Guide: https://datamcp.app/blog/agent-memory - Multi-agent memory architecture: https://datamcp.app/blog/multi-agent-memory - Claude memory comparison: https://datamcp.app/blog/claude-persistent-memory - Codex memory and MCP guide: https://datamcp.app/blog/codex-memory-mcp - Cursor memory, Memory Bank, and MCP guide: https://datamcp.app/blog/cursor-memory-mcp - Memory Bank MCP comparison: https://datamcp.app/blog/memory-bank-mcp ## Plans - Free: 1 source connection, 1 MCP link, 2 members, 7-day activity retention. - Pro: $19 per month, 3 source connections, 5 MCP links, 5 members, 30-day activity retention. - Enterprise: $49 per month, 15 source connections, 50 MCP links, 25 members, 365-day activity retention. - Agent Memory appends per user and memory connection: Free 20/min and 60/hour; Pro 120/min and 2,000/hour; Enterprise 600/min and 20,000/hour. - Agent Memory reads per user and memory connection: Free 120/min; Pro 600/min; Enterprise 3,000/min. - Agent Memory handoffs per user and memory connection: Free 60/min; Pro 300/min; Enterprise 1,000/min. - Agent Memory compactions per user and memory connection: Free 10/hour; Pro 100/hour; Enterprise 300/hour. - Additional Agent Memory connection-level and organization-level safety limits apply. Use the canonical pages above for current product details and limitations.