Stack Overflow MCP server icon

Stack Overflow

by Stack Overflow

HIPAA CompliantSOC2 ReadyISO 27001 Ready
Developer Tools3 tools

Search public Stack Overflow questions and pull full answer threads into an AI agent. Read-only, capped at 100 calls per user per day, and every application must visually attribute Stack Overflow as the source.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect Stack Overflow via MCP

https://mcp.stackoverflow.com

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Stack Overflow Tools & Capabilities (3)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • Attribution is a term of access, and no tool discharges it. Stack Overflow's MCP Terms require every application to visually indicate Stack Exchange as the source of retrieved content. There is no attribution tool, and no documented licence or canonical-URL field in either tool's response. Your prompt and your interface have to do it.
  • ShareAlike propagates onto anything you redistribute. Content is CC BY-SA 2.5, 3.0 or 4.0 depending on when it was posted. A derived document or dataset must carry the same licence; a conversational answer does not trigger it.
  • Caching, indexing and persistent storage are prohibited. Realtime access only. This is a direct constraint on conventional RAG designs.
  • Model training is prohibited without separate written permission, including by any IDE, extension or agent platform in the request path.
  • 100 calls per user per day, with no way to check the balance. No tool reports quota. The REST API's quota_remaining and backoff fields have no documented MCP equivalent.
  • Two tools are documented, three are listed. The extra name answers appears only in Anthropic's directory. We could not diff against a live handshake because the endpoint is gated, and Stack Overflow's GitHub repository holds a README with no source.
  • This is a beta server. Stack Overflow says so in its documentation and its README, and says additional tools may arrive in future phases. Neither the tool surface nor the rate limit should be treated as stable.
  • No OAuth scopes are discoverable. .well-known/oauth-protected-resource and .well-known/oauth-authorization-server both return 404 on the MCP host, and the 401 challenge carries a bare www-authenticate: Bearer with no resource_metadata pointer. The Stack Exchange REST API documents scopes including read_inbox, no_expiry, write_access and private_info, but we could not establish which, if any, the MCP consent screen requests. Nothing we found puts a boundary between reading and the API's write vocabulary at the OAuth grant — the boundary we can evidence is that no write tool is documented.
  • We could not read tool schemas or safety annotations. The endpoint returned 401 to an anonymous request, so no parameter-level detail is published here.
  • Search is lexical, not semantic. Stack Overflow describes so_search as lexical search over questions, so paraphrased queries match worse than verbatim error text.
  • Retrieved content is community-written and untrusted by construction. Anyone can post to Stack Overflow. Answers can be outdated, wrong or contested, and text arriving through the connector is a prompt-injection surface like any other user-generated corpus. Treat retrieved posts as data, never as instructions, and check code before running it.
  • stackoverflow.com/robots.txt carries Content-signal: search=no, ai-train=no with Disallow: /. The main site asks not to be crawled or used for AI training. The pages cited below are the vendor-nominated documentation URL and the legal documents it links to, fetched directly rather than crawled; we have restated their positions in our own words and linked rather than reproduced them.

Frequently asked questions

No. This connector reaches the public Stack Exchange Network only. Stack Overflow ships a separate MCP server for Stack Internal, formerly Stack Overflow for Teams, on a per-tenant endpoint of the form yourteam.stackenterprise.co/mcp. That one is a different product with a different address and a write surface this one does not have.

No. Both tools Stack Overflow documents are reads: one searches questions, one retrieves content by ID. The Stack Exchange REST API behind it does support posting, editing, deleting, voting and flagging, and none of that vocabulary appears in this connector. Stack Internal's MCP server does expose creation tools, which is the surface people confuse this with.

Attribution is mandatory and share-alike applies. Stack Overflow's MCP Terms require every application to visually indicate that the Stack Exchange Network is the source. User contributions are licensed CC BY-SA, so a redistributed adaptation must carry the same licence. For clients without web browsing, Stack Overflow says plain-text URLs suffice.

One hundred per user per day. Stack Overflow states the limit in both its API documentation and its MCP Terms of Use, and the Terms add that technical attempts to exceed or subvert it are treated as a willful breach that may lead to suspension, termination or legal action. No tool in the connector reports how much of the allowance remains.

No. Stack Overflow's MCP Terms restrict the service to realtime access and prohibit long-term or programmatic storage, indexing or caching of the content. The same clause forbids building anything that competes with or substitutes for the MCP Service or the Stack Overflow Network. This constrains RAG designs that would normally persist an embedding index.

No, not without separate written permission. Stack Overflow's MCP Terms prohibit using the service to train, fine-tune or improve machine-learning models, and extend that to any tool or platform that accesses the server. So an IDE, extension or agent in the path may not feed the retrieved content into its own model training either.

Anthropic's directory lists three and Stack Overflow documents two. The two Stack Overflow names are so_search and get_content. A third name, answers, appears only in the directory listing and in no Stack Overflow document we could find. Stack Overflow's page says additional tools may be added in future phases, so the surface is expected to move.

Stack Overflow calls it a beta server in three places. Its documentation, its GitHub README and its feedback request all use the word, and the README says feedback is critical to its evolution. The 100-call daily cap is framed as supporting prototyping and community projects rather than production traffic.

Sources

Use in Agentman

Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio

Server Info

Category
Developer Tools
Developer
Stack Overflow
Tools
3
Domain
mcp.stackoverflow.com

Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.