chatgpt-api/todo.md

## 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 <img height="..."> 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