mcp serverbeta

For AI agents

A hosted Model Context Protocol server over every tracked Salesforce Help article. JSON-RPC 2.0 over plain HTTP POST — no SSE, no session ids. Every tool is section-scoped and capped so it never floods your context window.

Beta: tools and response shapes can still change, and coverage is limited to the articles currently tracked. Pin nothing critical to it yet.

endpoint

POST https://releasehunter.com/mcp

No key, no signup, no auth header. Calls are rate limited to 20 a minute and 500 a day per caller.

client config

{
  "mcpServers": {
    "release-hunter": {
      "type": "http",
      "url": "https://releasehunter.com/mcp"
    }
  }
}

toolsearch_salesforce_docs(query, product_area?, limit?)

Search tracked Salesforce Help articles. Returns matching SECTIONS with headings and short excerpts, not full articles. Use this first to locate content, then get_doc_section for the full text.

toolget_doc_section(document_id, heading_path[])

Return the full text of one section, identified by its heading path from search_salesforce_docs. Prefer this over whole articles — Salesforce Help articles are long and will consume most of your context.

smoke test

curl -s https://releasehunter.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

response caps

Section excerpts cap at 800 characters, then “[truncated — call get_doc_section]”.
Default limit is 15 results, maximum 40.
Every item carries document_id, title, heading_path, source_url, captured_at, and freshness (fresh <24h | stale | hidden | removed).
Over the rate limit you get a normal tool error with the reset time, never a protocol error that drops the connection.