GitHub Discussions MCP server

How an agent would call it

Two steps: search returns refs, then data_fetch opens one of them. The arguments below are the arguments that were sent.

{
  "tool": "search",
  "arguments": {
    "query": "repo:vercel/next.js app router caching",
    "sources": [
      "github-discussions"
    ],
    "limit": 2,
    "runId": "<from run_begin>"
  }
}

{
  "tool": "data_fetch",
  "arguments": {
    "sourceId": "github-discussions",
    "ref": {
      "id": "vercel/next.js#98310"
    },
    "runId": "<from run_begin>"
  }
}

This runs against an account — send Authorization: Bearer <token> with the call. Get one at app.bedrocktap.com/dashboard/tokens; the free tier needs no card, and it reaches github-discussions like every other source.

What it costs. 0.45 credits per discussion returned, and 0.45 per fetch. The search above returned 2 discussions and cost 0.9 credits; the fetch cost 0.45.

What came back

{
  "ref": {
    "id": "vercel/next.js#98310",
    "url": "https://github.com/vercel/next.js/discussions/98310",
    "title": "How to stop Next.js from caching notFound() renders for invalid dynamic route params?"
  },
  "text": "How to stop Next.js from caching notFound() renders for invalid dynamic route params?\n\n### Summary\n\nTitle: notFound() results for bot-scanned invalid dynamic params get permanently cached, bloating disk — tried Cache Components, issue persists…",
  "fetchedAt": "2026-09-12T14:57:40.003Z"
}

Each ref carried repo, number, category, author, upvotes, comments, answered and createdAt in its meta — the two hits sat in the "Help" and "App Router" categories and neither had an accepted answer. The fetched payload’s text ran to 17,804 characters: the title, the body, the accepted answer where one exists, and up to twenty comments each prefixed with its author. Its data carried title, body, url, createdAt, upvoteCount, author, answer and comments.

Observed 2026-09-12. One call against a live source — ask again tomorrow and the rows move.

What a row contains

One row is one discussion: an owner/repo#number id, the discussion URL, and the title. meta carries the repository, the number, the category the maintainers filed it under, the author, the upvote count, the comment count, whether an answer has been accepted, and the creation time. The answered flag is the one to sort on — an accepted answer is usually the only part worth reading.

A question it answers. What are this project’s users still arguing about, and which questions has a maintainer actually settled?

Routing guidance

What an agent reads to decide whether github-discussions is the right source, and what to reach for when it is not.

Answers

Feature ideation, "how do I" threads, and maintainer answers marked as accepted. The forum half of GitHub, where `github-issues` is the defect half.

Query shape

GitHub search syntax — `repo:owner/name`, `label:x` — plus keywords.

Price

Priced in credits, per row, per query or per call depending on the source. Every call is quoted before it runs — `cost_estimate` returns the exact figure and spends nothing. See /pricing.

Latency

Fast (~1-3s).

Prefer instead

`github-issues` for bugs and concrete feature requests.

Cannot answer

Repos with Discussions disabled, private repos, or anything outside GitHub.

Caveat

Requires a GitHub token — the Discussions API has no unauthenticated access, and without one this returns nothing at all. Check health first.

Caveat

Sorting and reaction filters are applied after the fact, because upstream search is best-match only.

Related sources