This is a living document. We publish it so users and contributors can see where AnythingMCP is going and propose changes. Pull requests welcome.
For the current shipped state, see the latest release.
- More pre-built adapters — community-requested SaaS connectors prioritised in GitHub Discussions
- Improved adapter authoring DX — better validation errors, scaffolding CLI
- Tool-level rate limiting per role
- Sharper observability: per-tool latency histograms, error-rate dashboards
- Streaming tool responses — for long-running tool calls (DB queries, file generation)
- Tool composition — declaratively chain tools without leaving the gateway
- Multi-tenant Cloud parity — feature parity between self-hosted and cloud.anythingmcp.com
- Adapter marketplace — install community adapters from a curated index, similar to a plugin registry
- gRPC connector type — first-class support for gRPC services alongside REST/SOAP/GraphQL
- Webhook ingestion — register webhooks as MCP "events" consumable by clients
- Fine-grained tool versioning — pin AI agents to specific tool versions
- Native MCP elicitation & resources — beyond tools, full MCP spec coverage
- Tool A/B testing — version-flag tool definitions and route per-user
- Self-tuning rate limits based on AI client behavior
- OpenTelemetry exporter for traces and metrics
- Plugin SDK in TypeScript and Python for custom auth/transformations
- Building our own LLM client
- Code-execution sandbox (use E2B or similar instead)
- Replacing Postman/Insomnia as a general API client
Open a feature request or start a thread in Discussions → Ideas.