docs(TSP-1356): document MCP run-only access and Viewer role MCP support

[claude]

Summary

• MCP server page (integrations/mcp/mcp-server.mdx): Expanded the Authentication section to document the OAuth consent flow and the new Run-only access toggle. Added a Roles & access levels table showing that Viewer-role users can now connect via MCP (locked to run-only) while Member/Editor/Admin users default to full access but can opt into run-only. Updated the FAQ answer for "Can I restrict which tools are available via MCP?" to mention the toggle alongside project separation.
• RBAC page (enterprise/rbac.mdx): Updated the project-level permissions table to show Viewers can mint a personal API key (✅†) for MCP OAuth connections. Added a footnote Note explaining this is MCP-specific and that Viewers connecting via MCP are automatically placed in run-only mode. Updated the Viewer role description to mention MCP run-only execution.

Context

Implements documentation for PR #16022, which introduces:

1. Viewer-role users can now mint their own API keys and connect MCP clients via the OAuth consent flow.
2. A new "Run-only access" toggle on the MCP OAuth consent page — write and delete tools are entirely absent from the MCP tool list in run-only mode. Viewer-role users are locked to run-only; higher-privilege users can opt in voluntarily.

Linear: https://linear.app/relevance/issue/TSP-1356/

[linear]

TSP-1356

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
relevanceai	🟢 Ready	View Preview	Jul 21, 2026, 7:09 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
relevanceai	🟡 Building	–	Jul 21, 2026, 7:08 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[github-actions]

🎯 Vibe check

Reviewed: 2 files (2 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	Two heading case violations, two banned-word accordion titles, and a cluster of uncapitalized product terms scattered across mcp-server.mdx (agents/tools/workforces/knowledge).
🟡	Technical clarity	7/10	"Write and delete tools are entirely absent" is ambiguous (tools the product or capabilities?). "Directly without agents" in the Chat role accordion is genuinely confusing. Temporal "now" language in a <Note> will go stale.
🟢	Non-technical clarity	9/10	Both pages are well-structured and explain concepts before instructions. One accordion on rbac.mdx has a sentence that will lose a non-technical reader.
🟢	Structure	9/10	<Info> block in rbac.mdx contains a **Note:** bold label, which CLAUDE.md explicitly forbids inside callouts. Everything else is well-organized.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The core content is solid — the new Viewer MCP access and run-only toggle are documented accurately and the two files are consistent with each other. The main cleanup is mechanical: sentence case on two headings, retiring two accordion titles that use the banned word "powerful", and capitalizing Relevance AI product terms (Agents, Tools, Workforces, Knowledge) throughout mcp-server.mdx.

🔧 Issues (13)

enterprise/rbac.mdx

• enterprise/rbac.mdx:22 — ## Best Practices → ## Best practices (sentence case; "Practices" is not a proper noun)
• enterprise/rbac.mdx:149 — <Accordion title="Controlled agent visibility"> → "Controlled Agent visibility" — "agent" here refers to a Relevance AI Agent (product term)
• enterprise/rbac.mdx:161 — "Can have conversations with LLMs and in-built Chat Agents directly without agents." — "directly without agents" is ambiguous. Probably means "without needing a custom Agent configured for Chat". Suggest: "Can have LLM conversations and use in-built Chat Agents without a custom Agent."
• enterprise/rbac.mdx:165 — <Accordion title="More powerful than Viewer"> — "powerful" is a banned word per CLAUDE.md. Suggest: "More capable than Viewer" or "Can do more than Viewer"
• enterprise/rbac.mdx:165 (content) — "Can trigger Chat and interact with agents" → "interact with Agents"
• enterprise/rbac.mdx:169 — <Accordion title="Less powerful than Member"> — "powerful" is a banned word. Suggest: "Less capable than Member" or "Cannot do everything Member can"
• enterprise/rbac.mdx:217 — ## Workforce Permissions → ## Workforce permissions (sentence case)
• enterprise/rbac.mdx:235–237 — <Info> block starts with **Note:**. CLAUDE.md says callouts must contain no bold labels. Remove the **Note:** prefix — the <Info> component already signals this is a note.

integrations/mcp/mcp-server.mdx

• integrations/mcp/mcp-server.mdx:8 — "gives any MCP-compatible AI client direct access to your agents, tools, and knowledge" → "Agents, Tools, and Knowledge" (Relevance AI product terms)
• integrations/mcp/mcp-server.mdx:181 — "view and execute agents, tools, and workforces" → "Agents, Tools, and Workforces" (product terms). Also on the same line: "write and delete tools are entirely absent from the MCP tool list" — "tools" here is ambiguous (it reads as if specific tools named "write" and "delete" are removed). Suggest: "write and delete operations are entirely absent from the MCP tool list".
• integrations/mcp/mcp-server.mdx:195 — "Viewer-role users can now connect to Relevance AI via MCP. Previously, only Editor-role and above could connect." — temporal language ("now", "Previously") will go stale. Rewrite as evergreen: "Viewer-role users can connect to Relevance AI via MCP in run-only mode."
• integrations/mcp/mcp-server.mdx:259 — "call Relevance AI tools" → "call Relevance AI Tools"
• integrations/mcp/mcp-server.mdx:264,271,275 — "triggering agents via MCP", "agents that complete quickly", "agents with workforce nodes" → "Agents" in all three (these are Relevance AI Agents). Note that line 324 on the same page already uses the correct "Agents" — making this an internal inconsistency.

🏗️ Page structure (1)

• enterprise/rbac.mdx:165–171 — The two accordion items "More powerful than Viewer" and "Less powerful than Member" each contain a single sentence and exist only to describe relative positioning of the Chat role. This level of hierarchy adds accordion interaction overhead for content that reads naturally as two bullet points within the Chat role's intro paragraph. Consider collapsing these into prose: "The Chat role sits between Viewer and Member — it can trigger Chat and interact with Agents (with proper asset permissions), but cannot create assets or access the web app."

⚠️ Contradictions (0)

No contradictions found. The Viewer MCP run-only access is described consistently between both files: rbac.mdx line 100 and 137 match the table and explanatory text in mcp-server.mdx lines 187–192.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~1,060

Files read: enterprise/rbac.mdx (380 lines), integrations/mcp/mcp-server.mdx (357 lines), enterprise/user-level-authentication.mdx (220 lines), integrations/mcp/mcp-client.mdx (103 lines)