Refinry reviews read by the tools you already use. Two audiences below: a plain-language guide for people, and a machine-readable block for AI agents.
Refinry publishes evidence-cited reviews of open-source projects for ministry use. MCP (Model Context Protocol) lets a compatible AI assistant read live data from outside services.
Refinry is a hosted service for evidence-cited open-source reviews, not the unrelated product Refine. No Cargo, Rust daemon, OS service, or local checkout is needed to connect to the hosted service.
Use a client that supports remote MCP over Streamable HTTP. Availability depends on your client, account, and workspace permissions. Ordinary chat can explain setup; it does not automatically install or configure a connection.
MCP server URL: https://open-serve-mcp.vercel.app/mcp/ — use this address, not the /docs or /docs.md documentation URL.
For ChatGPT on the web, official guidance describes Settings → Apps → Advanced settings → Developer mode, then Create app for remote MCP. Labels and eligibility vary; workspace permissions may apply, and the flow may include Scan Tools/Create. The server probes do not establish compatibility with your ChatGPT account; the actual ChatGPT UI is untested.
Official developer-mode guidance · Workspace and setup guidance
If connection controls are unavailable, ask your workspace administrator or use another remote-MCP-capable client. Do not substitute a local installation.
Every Refinry review page ends with a "Start building" section: a ready-made connection prompt, and a first-step prompt you can shape to your situation. Open any repo's review and scroll to the bottom — or start from the leaderboard. If the connection fails there, the steps on this page cover the manual path.
Tool names and logos belong to their owners; shown so you can pick your own client. Not affiliated.
The block below is the machine-readable reference. It is also served as raw markdown at /docs.md.
# Refinry MCP server — agent referenceProduct: Refinry, the hosted open-source review service; not Refine or a local daemon.MCP server identity: serverInfo.name=refinry; title=Refinry.Endpoint: https://open-serve-mcp.vercel.app/mcp/Transport: Streamable HTTPDocs (human guide, not a server URL): https://refinry.dev/docsFull markdown guide containing this reference (not a server URL): https://refinry.dev/docs.mdAuthentication: public anonymous access requires no credential.## Connection boundariesUse the endpoint above only in a client that supports remote MCP. Availability depends on client capability, account eligibility, and workspace permissions. Ordinary chat may guide setup but cannot be assumed to configure connections. Do not suggest Rust, cargo, a local daemon, an OS service, or machine changes for connecting to Refinry. Confirm access only after an actual successful tool call. Successful server probes do not establish ChatGPT account compatibility; its UI has not been tested.## ToolsNotation: ? means optional; null is accepted only where shown. Defaults and authoritative schemas are available through tools/list.- search_repos(q?: string | null, category?: string | null, license_spdx?: string[] | null, self_hosted?: boolean | null, min_mri?: number | null, limit?: integer = 10, offset?: integer = 0) -> scored repository search- get_repo_review(full_name: string) -> evidence-cited review, dimension scores, confidence, safety tags, provenance and audit trail pointer- top_by_category(category?: string = "ALL", limit?: integer = 10) -> leaderboard with security-first tie-breaking- ministry_fit_check(use_case: string, constraints?: object | null, max_recommendations?: integer = 3) -> scored-fit recommendations with reasons, evidence and confidence## Example calls- search_repos q="church management"- get_repo_review full_name="example-owner/example-repo"- top_by_category category="ALL" limit=10- ministry_fit_check use_case="Volunteer check-in for our ministry"## Rate limitsAdvertised anonymous limit: 60 requests/minute, burst 10. The server also advertises 300 requests/minute with a free key; no key is needed for anonymous access. On HTTP 429, honor Retry-After if supplied; otherwise back off before retrying. Health endpoint: GET https://open-serve-mcp.vercel.app/api/v1/health## Citing reviewsAlways cite the review date (reviewed_at) and the Trust Score together when quoting a Refinry review. If either is absent, state that limitation. Do not present an old review as a fresh assessment or invent score-decay calculations.
The markdown connection guide and agent reference served at /docs.md; the client table above is additional HTML-page guidance:
# Refinry Docs — connecting your AI assistant ## What is this MCP thing? Refinry publishes evidence-cited reviews of open-source projects for ministry use. MCP (Model Context Protocol) lets a compatible AI assistant read live data from outside services. Refinry is a hosted service for evidence-cited open-source reviews, not the unrelated product Refine. No Cargo, Rust daemon, OS service, or local checkout is needed to connect to the hosted service. Use a client that supports remote MCP over Streamable HTTP. Availability depends on your client, account, and workspace permissions. Ordinary chat can explain setup; it does not automatically install or configure a connection. Docs URL: https://refinry.dev/docs Raw docs URL: https://refinry.dev/docs.md MCP server URL: https://open-serve-mcp.vercel.app/mcp/ ## Connect Refinry (step by step) 1. Check that your client and account allow remote MCP connections; workspace approval may be required. 2. Open the client’s connection settings and add a remote MCP server or app; labels vary. 3. Enter the server URL: https://open-serve-mcp.vercel.app/mcp/ — not the /docs or /docs.md documentation URL. 4. Use No Authentication for public anonymous access, where the client offers that option. 5. Save and complete any tool scan or approval. Run search_repos with q="church management"; only a successful tool result confirms access. ### ChatGPT For ChatGPT on the web, official guidance describes Settings → Apps → Advanced settings → Developer mode, then Create app for remote MCP. Labels and eligibility vary; workspace permissions may apply, and the flow may include Scan Tools/Create. The server probes do not establish compatibility with your ChatGPT account; the actual ChatGPT UI is untested. Official guidance: https://developers.openai.com/api/docs/guides/developer-mode Workspace and setup guidance: https://help.openai.com/articles/12584461 If connection controls are unavailable, ask your workspace administrator or use another remote-MCP-capable client. Do not substitute a local installation. ## Where do I start with a specific project? Open a repo review and scroll to "Start building" for a connection-help prompt and a first-step prompt. Pasting a prompt requests guidance; it does not create a connection automatically. <!-- REFINRY-MCP-BEGIN --> # Refinry MCP server — agent reference Product: Refinry, the hosted open-source review service; not Refine or a local daemon. MCP server identity: serverInfo.name=refinry; title=Refinry. Endpoint: https://open-serve-mcp.vercel.app/mcp/ Transport: Streamable HTTP Docs (human guide, not a server URL): https://refinry.dev/docs Full markdown guide containing this reference (not a server URL): https://refinry.dev/docs.md Authentication: public anonymous access requires no credential. ## Connection boundaries Use the endpoint above only in a client that supports remote MCP. Availability depends on client capability, account eligibility, and workspace permissions. Ordinary chat may guide setup but cannot be assumed to configure connections. Do not suggest Rust, cargo, a local daemon, an OS service, or machine changes for connecting to Refinry. Confirm access only after an actual successful tool call. Successful server probes do not establish ChatGPT account compatibility; its UI has not been tested. ## Tools Notation: ? means optional; null is accepted only where shown. Defaults and authoritative schemas are available through tools/list. - search_repos(q?: string | null, category?: string | null, license_spdx?: string[] | null, self_hosted?: boolean | null, min_mri?: number | null, limit?: integer = 10, offset?: integer = 0) -> scored repository search - get_repo_review(full_name: string) -> evidence-cited review, dimension scores, confidence, safety tags, provenance and audit trail pointer - top_by_category(category?: string = "ALL", limit?: integer = 10) -> leaderboard with security-first tie-breaking - ministry_fit_check(use_case: string, constraints?: object | null, max_recommendations?: integer = 3) -> scored-fit recommendations with reasons, evidence and confidence ## Example calls - search_repos q="church management" - get_repo_review full_name="example-owner/example-repo" - top_by_category category="ALL" limit=10 - ministry_fit_check use_case="Volunteer check-in for our ministry" ## Rate limits Advertised anonymous limit: 60 requests/minute, burst 10. The server also advertises 300 requests/minute with a free key; no key is needed for anonymous access. On HTTP 429, honor Retry-After if supplied; otherwise back off before retrying. Health endpoint: GET https://open-serve-mcp.vercel.app/api/v1/health ## Citing reviews Always cite the review date (reviewed_at) and the Trust Score together when quoting a Refinry review. If either is absent, state that limitation. Do not present an old review as a fresh assessment or invent score-decay calculations. <!-- REFINRY-MCP-END -->
