Skip to content

chore: review and expand bundled provider catalog — MCP providers, correct type classification, JSONC migration #408

Description

@manojbajaj95

Summary

The bundled provider catalog needs a systematic review pass before the v2 release:

  1. Add MCP providers — Anthropic publishes a list of remote MCP servers (e.g. Brave Search, Cloudflare, Filesystem, GitHub, GitLab, Google Maps, Memory, PostgreSQL, Puppeteer, Sequential Thinking, Slack, Stripe, Linear, Notion, Sentry, etc.) that agents commonly connect to. These should be first-class providers with type: mcp so users can authsome login <mcp-provider> and get credentials injected into their MCP client config.

  2. Fix type misclassifications — The type field was added recently but the audit is incomplete:

    • 71 of 73 providers are typed "app" — several are misclassified:
      • linkedin-browser, x-browserauth_type: browser but type: app → should be browser
      • openai, google-vertex-ai — already llm
      • anthropic (not yet present), mistral, cohere, groq, together-ai, fireworks, perplexity, etc. — missing LLM providers that should have type: llm
    • Define and document what each type means:
      • llm — an LLM/AI inference API
      • app — a SaaS product / REST API (default)
      • mcp — exposes an MCP server endpoint; credential becomes the bearer token for the MCP URL
      • browser — cookie/session extracted from a real browser session
  3. Migrate all provider configs to JSONC — JSONC support was added in feat: add ProviderType classification and JSONC support for providers #406. Rename every .json file to .jsonc and add inline comments documenting:

    • Why non-standard scopes were chosen
    • Notes on quirky token endpoints or refresh behaviour
    • Which export env vars agents should prefer

Acceptance criteria

  • New mcp providers added (at minimum: Brave Search, Stripe, Sentry, Linear MCP, Notion MCP, GitHub MCP, Slack MCP, Cloudflare, PostgreSQL)
  • linkedin-browser and x-browser reclassified to type: browser
  • Missing LLM providers added (Anthropic, Mistral, Groq, Together AI, Fireworks, Perplexity) with type: llm
  • All 73+ provider files renamed from .json.jsonc
  • Each .jsonc file has at least a top-level comment block explaining the provider and any non-obvious config choices
  • ProviderType enum in the Python model is updated if any new type values are introduced
  • uv run pytest passes, uv run ruff check, uv run ty check clean

Reference

Metadata

Metadata

Assignees

Labels

enhancementNew feature or requestready-for-agentFully specified, ready for an AFK agent

Type

No type

Fields

No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions