Skip to content

Releases: deghosal-2026/mcp-fabric

v0.2.0 — Resource-Aware Policy

Choose a tag to compare

@deghosal-2026 deghosal-2026 released this 25 Jul 20:32

Resource-Aware Policy (Dynamic Resource Dimensions)

The OPA policy engine now evaluates agent_class + capability + (resource dimension) triples instead of just agent_class + capability. Platform engineers define resource dimensions per-capability (env, tenant, service), bind allowed values to agent identities and packs, and OPA checks every request against those bindings at runtime.

What's included

  • 4 new DB tables — resource_dimensions, dimension_value_map, identity_resource_bindings, pack_resource_bindings
  • 10 new API endpoints — dimension CRUD, identity/pack resource binding management
  • OPA Rego extension — resource_allowed gate, dim_allowed helper, resource_violations set
  • Admin UI — dimension modals on capabilities, binding modals on agent classes and packs, resource violation display in approvals
  • Audit — resource_check captured in every capability request audit event, filterable via resource_violation=true
  • 47 new tests — 13 backend integration, 12 OPA Rego, 2 merge_bindings, 3 Playwright E2E, 7 UI test case specs

Documentation

v0.1.0 — MCP Fabric Initial OSS Release

Choose a tag to compare

@deghosal-2026 deghosal-2026 released this 25 Jul 12:37

A control plane for AI agents — register MCP servers, normalize tools into capabilities, control access with OPA policies, require human approval for sensitive actions, and audit everything.

Highlights:

  • Server registry with auto-discovery of tools
  • Capability catalog with domain:action normalization
  • OPA Rego policy engine for access governance
  • Agent classes with identity token management
  • Capability packs — bundle and assign to agent classes
  • Human-in-the-loop approvals for sensitive capabilities
  • Audit log with event filtering and JSON export
  • Trust posture dashboard with per-class trust levels
  • Admin user management with RBAC and MFA
  • 474 automated tests (backend, UI, OPA, E2E)
  • Published on PyPI as mcp-fabric-toolmesh

See CHANGELOG and ROADMAP for full details.