thetahealth/mirobody

git clone https://github.com/thetahealth/mirobody.git && cd mirobody && ./deploy.sh

Usage

• •Deploy via Docker (./deploy.sh generates .env and starts Postgres/Redis/Mirobody), then open http://localhost:18080 (demo login demo1@mirobody.ai / 777777).
• •Drop Python tools into tools/ or SKILL.md skills into skills/ — auto-discovered as MCP tools with zero config.
• •Connect MCP clients via universal-mcp-proxy to http://localhost:18080/mcp, or set MCP_PUBLIC_URL in config for remote HTTPS access.
• •OPENROUTER_API_KEY is required for the Deep agent; code execution needs E2B_API_KEY.

Key features

• ✓Works simultaneously as an MCP client and an OAuth-enabled MCP server with local and HTTP remote access.
• ✓Three agent types: DeepAgent (full MCP tool support), MixAgent (two-phase model fusion), BaseAgent (direct LLM conversation).
• ✓FHIR & health standards: 400+ indicators with LOINC/SNOMED CT/RxNorm search; Pulse pipeline ingests from 300+ wearables, Apple Health, and EHR.
• ✓Claude Agent Skills (SKILL.md + metadata.json) support with auto-discovery.
• ✓PostgreSQL-backed file tools and E2B cloud-sandboxed code execution with auditability.

Best for

Personal-intelligence apps — health/finance analysis over your own data, self-hosted, exposed as MCP tools.

Caveats

• ⚠Requires Docker, Docker Compose, and Git LFS; MixAgent is experimental.
• ⚠Deep agent needs OPENROUTER_API_KEY; the execute tool needs E2B_API_KEY.
• ⚠Full Agent Skills spec (scripts/, references/, assets/, allowed-tools enforcement) is planned but not yet supported; metadata.json is a Mirobody-specific requirement.
• ⚠Sensitive config is encrypted with CONFIG_ENCRYPTION_KEY; API keys are auto-encrypted on load.