Where MCP Servers Get Listed in 2026
There are four places an MCP server is normally listed in 2026, and three of them are filled by the person who wrote the server. On two of the awesome-lists and on the official registry, an entry exists because its own author put it there. That is not a criticism of any of them — it is the fastest way to build a broad, current list, and it is why those listings are far larger than this one. But it means the answer to “is this thing any good” is not in the fact that it is listed, and the pages themselves rarely say so. This page states each listing’s submission policy in its maintainers’ own terms, so you know what a listing is evidence of before you rely on it. Every figure and quotation below was checked against the primary source on 5 August 2026.
The four listings, side by side
Each row is that listing’s own stated policy, not an assessment of it. “Who can add an entry” is the column that changes what a listing means:
| Listing | How an entry gets in | Who can add one | Size | What is checked |
|---|---|---|---|---|
| punkpeye/awesome-mcp-servers | Fork the repo, edit README.md, open a pull request | Anyone, including the server’s own author | 3,387 linked entries | Format and category, by review of the pull request |
| wong2/awesome-mcp-servers | A submission form at mcpservers.org; pull requests are not accepted | Anyone, including the server’s own author | 537 linked entries | Whatever the site’s maintainer applies before publishing |
| The official MCP registry | The maintainer publishes the server themselves, through the registry’s publisher tool | The server’s own maintainer, who must prove they control its namespace | A live public API, paginated; still labelled preview | Namespace ownership and schema validity |
| Agentwire | There is no submission route | Nobody — entries are found, never submitted | 75 entries | That two independent people already vouched, in two different ways |
punkpeye/awesome-mcp-servers takes pull requests, including from agents
An entry joins punkpeye/awesome-mcp-servers when somebody forks the repository, edits README.md and opens a pull request that is merged. The list’s CONTRIBUTING.md sets out those steps and asks that a new server arrive with its name linked to its repository, a short description, and placement under the right category. Nothing in that process is a check on whether the server works — it is a check that the entry is formatted and filed correctly, which is what a hand-kept list of this size can realistically do. On 5 August 2026 the README carried 3,387 linked entries and the repository had 2,489 open issues.
The list also publishes an explicit route for automated contributors. Its CONTRIBUTING.md says: “If you are an automated agent, we have a streamlined process for merging agent PRs. Just add 🤖🤖🤖 to the end of the PR title to opt-in. Merging your PR will be fast-tracked.” That is a deliberate policy by a maintainer keeping pace with a fast-moving ecosystem, and it is in routine use — of the five most recently opened pull requests on 5 August 2026, two carried that marker in their titles. It is worth knowing when you read the list, because it means a listed server may have been added by software on its author’s behalf.
wong2/awesome-mcp-servers does not take pull requests at all
wong2/awesome-mcp-servers is filled through a form rather than through Git. The note at the very top of its README reads: “We do not accept PRs. Please submit your MCP on the website: https://mcpservers.org/submit”. Submissions therefore pass through whatever the maintainer of that site applies before they appear, which is a different and more centralised gate than a merged pull request, and it is a plausible part of why the list is smaller and moves more slowly: 537 linked entries on 5 August 2026, last updated on 13 July 2026, against a repository that had 36 open issues.
The official MCP registry is self-publish, and still in preview
The registry run by the Model Context Protocol project describes itself in its README as providing “MCP clients with a list of MCP servers, like an app store for MCP servers”. Servers are published by their own maintainers through the registry’s publisher tooling, and the thing the registry verifies is namespace ownership — that whoever is publishing controls the name they are publishing under. That is a real and useful guarantee, and it is a guarantee about identity rather than about quality. Its public API answered on 5 August 2026 at registry.modelcontextprotocol.io/v0/servers, returning servers in pages of thirty, each carrying an official status field. That page size was a hundred when this page was first written two days earlier, which is its own small argument for reading the API rather than a description of it.
One thing to carry across from its README: the registry launched in preview on 8 September 2025 and entered an API freeze at v0.1 on 24 October 2025, with general availability still to come. Its own documentation warns that breaking changes or data resets may occur. If you are building against it rather than browsing it, that is the sentence that matters.
Agentwire has no submission route, and is small because of it
Agentwire is the fourth row and the smallest by a wide margin: 75 entries today. There is no form, no pull request and no way for the author of a tool to add their own work, which is the entire reason the index is that size. An entry appears only when at least two independent affirmations already existed — a maintainer put it on a list they publish, a Hacker News thread pointed at it, it is new and already past a star floor, or its own repo metadata carries two or more agentic topics. Two, never one, so that a coincidence on any single signal does not qualify.
The trade is not subtle and it runs in both directions. A listing anyone can add to will always be broader, more current and more useful for “does an MCP server exist for X”. An index nobody can add to will always be narrower, and is only worth reading for the different question of what somebody else already staked their name on. How eligibility is decided has the thresholds, and where the entries come from names the specific lists and threads behind the entries currently in the index.
Which listing answers which question
If the question is “is there an MCP server for this API”, the broad listings answer it and this one usually will not: start with punkpeye/awesome-mcp-servers for coverage, or the official registry if you need something machine-readable with a verified namespace. If the question is “which of these did somebody independent actually vouch for”, that is what a two-signal index is for, and it is a much shorter answer. If the question is “does this server work, and is it safe to point at my data”, no listing on this page answers it, including this one.
What none of these listings tell you
None of the four tests the servers it lists. Not the awesome-lists, not the official registry, and not Agentwire. A listing is a record that something exists and, at most, that somebody was willing to file it under their own name — it is not a security review, a maintenance guarantee or a statement that the code runs. An MCP server is something you are giving access to a model and often to your data, and every one of these listings leaves that judgement with you.
How this page is kept honest
Every number and every quotation above was read from the primary source on 5 August 2026 — the two lists’ own README and CONTRIBUTING files, the registry’s README, and the registry’s live API — and each is recorded with its source in this site’s claim register so it can be re-checked rather than trusted. These facts decay: a list can change its submission policy in a single commit, and the registry will leave preview at some point. If something here is out of date or misstates a policy you maintain, say so at hello@kynth.studio and it gets corrected.
This page ranks nothing. The four listings are built by different people to answer different questions, and the differences described above are their stated policies rather than anybody’s score.