## API Gateway - **MCP**: `https://gateway.agentic.so/@{username}/{project-slug}/mcp` - **HTTP**: `GET/POST` `https://gateway.agentic.so/@{username}/{project-slug}/{tool-name}` ## TODO: MVP - **website** - stripe - double check stripe upgrade flow and add fireworks - improve upgrade flow UX - should we bypass stripe for `free` plans to increase conversions? - **replace json pricing plans and consumers with actual designs** - double-check free-tier rate-limits for `@agentic/search` - docs: add notes about constraints on mcp origin servers (static tools) - **api keys should go beyond 1:1 consumers** - **currently not obvious how to get api key** - marketplace public project detail page - add breadcrumb nav: marketplace > @agentic > search - add last published date somewhere - tool input/output schemas; move `$schema` to the top - break out into a few subcomponents; some can be server components - mcp inspector - improve private project page - link to public page if published - list deployment versions - **create agentic products for select legacy tools** - marketplace index page - add disclaimer about public beta - add search / sorting - replace render for api and/or add turbo for caching (too slow to deploy) - create slack + notifications - consider changing homepage hero CTA to include publishing ## TODO: Post-MVP - **API gateway** - oauth flow - https://docs.scalekit.com/guides/mcp/oauth - custom oauth flow might need to use separate domains per project instead of separate pathnames? - since the .well-known routes and standard oauth 2.1 routes are all at the top-level? - **e2e tests for usage tracking and reporting** - marketplace - may need a different flag besides `private` for inclusion on the marketplace - projects may be public but not accepted into the marketplace? - => **punt on this for mvp** - **website** - llms.txt and llms.txt for all projects - handle browser back/forward with `?next=` - add some social proof to signup page - first-party deployment hosting - api gateway stress tests - auth - custom auth provider configs for projects/deployments - stripe - re-add coupons - declarative json-based pricing - like Saasify and https://github.com/tierrun/tier - https://github.com/tierrun/tier/blob/main/pricing/schema.json - https://blog.tier.run/tier-hello-world-demo - stripe connect - stripe-related billing emails - re-add support for teams / organizations - consider switching to [consola](https://github.com/unjs/consola) for logging? - consider switching to `bun` (for `--hot` reloading!!) - validate stability of pricing plan slugs across deployments - same for pricing plan line-items - replace `ms` package - **API gateway** - **do we just ditch the public REST interface and focus on MCP?** - SSE support? (no; post-mvp if at all; only support [streamable http](https://modelcontextprotocol.io/specification/2025-03-26/basic/transports#streamable-http) like smithery does, or maybe support both?) - signed requests - add support for custom headers on responses - add ability to only report stripe usage on non-cached requests - add support for ToolConfig.cost defaulting to 1, to easily support tools which cost multiple "credits" - extra `Sentry` instrumentation (`setUser`, `captureMessage`, etc) - test handling of binary bodies and responses - improve logger vs console for non-hono path and util methods - consider replacing `eventId` with uuids for `requestId`? - openapi origin kitchen sink - add more test cases to e2e tests for diff content types - mcp origin kitchen sink - add support for `/sse`? need to test with claude desktop - test handling of resources - `@agentic/platform-hono` - fix sentry middleware - https://github.com/honojs/middleware/blob/main/packages/sentry/src/index.ts - https://github.com/honojs/middleware/issues/943 - https://github.com/getsentry/sentry-javascript/tree/master/packages/cloudflare - **Origin MCP servers** - how to guarantee that the request is coming from agentic? - `_meta` for tool calls - _still need a way of doing this for initial connection requests_ - _=> ask in the official mcp developers discord_ - mcp auth provider support - test binary bodies / responses / mcp resources - resources - prompts - other MCP features? - additional transactional emails - handle or validate against dynamic MCP origin tools - upgrade to zod v4 - decide whether deployment fields like `defaultRateLimit` and others should be generated and stored in the db, or should be inferred based on `undefined` values - support multiple rate-limits by slug - RateLimit-Policy: "burst";q=100;w=60,"daily";q=1000;w=86400 - https://datatracker.ietf.org/doc/draft-ietf-httpapi-ratelimit-headers/ - handle hosting of deployment and user images - about page inspiration: https://mastra.ai/about - simplify overlap between `@agentic/core` and `@agentic/platform-core` - consider simplifying `AgenticToolClient` to only require one package per TS LLM SDK - `createAISDKToolsFromIdentifier(projectIdentifier)` - consider allowing deployments to specify their proxy secret instead of a single, per-project proxy secret - add docs on using multiple tools with `AIFunctionSet` - ensure all agentic tool inputSchemas support openai strict mode by default - openapi tool extraction currently supports optional properties - mcp tool inputSchemas may not support openai strict mode either - maybe default `strict` to `false` in `createAIFunction` for now? - also add `@agentic/json-schema` to `createJsonSchema` parsing instead of current no-op - add support for [`@google/genai`](https://github.com/googleapis/js-genai) tools adapter - currently difficult due to their use of non-standard json schemas - validate example args against the tool's input schema during config validation - add scroll appearance motion to hero animation - add ts sdk examples to e2e tests - add feature about optimized context to docs - import react example usage component into docs - add [ping](https://modelcontextprotocol.io/specification/2025-03-26/basic/utilities/ping) support to mcp servers - basic account page on website - edit name, profile photo, etc - **public project detail page metadata** - fix readme css not taking effect because of tailwind css preflight which sets `img, video { height: auto }` - we still want this for every other scenario; just want to sandbox the github-style readme markdown css... - add `--llm` flag to cli (https://x.com/badlogicgames/status/1940370344990441726) - support claude Desktop Extensions ([DXT](https://github.com/anthropics/dxt)) ([post](https://www.anthropic.com/engineering/desktop-extensions)) - add mcp-remote to support stdio-only mcp clients like claude desktop - add docs on how to use with [chatgpt's mcp custom connectors](https://platform.openai.com/docs/mcp) ([requires pro account or team/enterprise/edu workspace](https://help.openai.com/en/articles/11487775-connectors-in-chatgpt#h_d2a53d4230)) - add support for enterprise / custom / contact us pricing