docs(TSP-1335): document evaluation version picker and MCP tool

[claude]

Summary

• Version picker in run modals: Added a new step in the "Running evaluations" section of the Evals page documenting the Version dropdown, which lets users select a specific historical or draft version to evaluate (applies to both agents and workforces). Defaults to the current active version.
• Run Summary table: Updated the "Agent Version" row description to reflect that it shows the version selected in the run modal.
• FAQ update: Updated "Can I rerun a previous evaluation?" to mention that users can target specific versions via the Version dropdown.
• MCP tool documentation: Added a new "Running evaluations via MCP" subsection documenting the relevance_run_evaluation tool with its agent_id, test_set_id, and optional version_id parameters.
• Workforce bug fix note: Added a note that workforce evaluations now correctly run against the selected version (previously always used active published version).
• MCP server page: Added an "Available tools" table listing relevance_trigger_agent, relevance_trigger_agent_async, relevance_poll_agent_result, and relevance_run_evaluation with brief descriptions.

Linear issue

https://linear.app/relevance/issue/TSP-1335/

🤖 Generated with Claude Code

[linear]

TSP-1335

[mintlify]

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

Project	Status	Preview	Updated (UTC)
relevanceai	🟢 Ready	View Preview	Jul 10, 2026, 5:03 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	Product terms "agent," "tool," and "knowledge" appear lowercase in 10+ spots across both files where they refer to the Relevance AI product features (Agent, Tool, Knowledge). Four accordion type-name titles also violate sentence case. Bold label inside a callout.
🟡	Technical clarity	8/10	mcp-server.mdx says "exposes the following tools" and lists 4, but the agent-skills repo documents 46 MCP tools — the "following" framing reads as a complete list and may mislead. The changelog Note in evals.mdx also muddies what the current behavior is.
🟢	Non-technical clarity	9/10	Both pages are well organized with FAQs, worked examples, and step-by-step instructions. Minor: "scenario" is used in the evals intro before being defined.
🟡	Structure	8/10	evals.mdx has a changelog-style sentence in a <Note> that records past (fixed) behavior — this will become noise. The best practices section uses <CardGroup> for what are essentially advisory tips.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: Both pages are thorough and technically accurate — the evals page in particular is a genuinely useful reference with good examples and a well-structured cost breakdown. The main drag on the PR is a recurring capitalization pattern: "agent," "tool," and "knowledge" go lowercase in prose and table cells throughout both files, even when they clearly refer to Relevance AI product features. A focused find-and-replace pass will close most of the gap.

🔧 Issues (13)

evals.mdx — product term capitalization

• evals.mdx:7 — **Rollout Status**: is a bold label inside an <Info> callout — CLAUDE.md prohibits bold labels inside callouts. Remove the label; start with: "Evals is rolling out progressively, starting with Enterprise customers…"
• evals.mdx:11 — "sub-agents and tools inside a Workforce" → "sub-agents and Tools inside a Workforce" (Tools = Relevance AI product feature)
• evals.mdx:93 — "a sub-agent or tool in the Workforce" → "Tool"; "a particular sub-agent used a given tool" → "a given Tool"
• evals.mdx:350 — "the agent's own run" → "the Agent's own run"
• evals.mdx:351 — "the agent's own tool runs" → "the Agent's own Tool runs"
• evals.mdx:354 — "the agent's own run" → "the Agent's own run"
• evals.mdx:360 — "The agent's own run during the test" → "The Agent's own run during the test"; "each tool the agent runs" → "each Tool the Agent runs"

evals.mdx — accordion title sentence case

• evals.mdx:58 — Accordion title "LLM Judge" → "LLM judge" (LLM is an acronym; "judge" is not a proper noun)

• evals.mdx:68 — Accordion title "Text Includes" → "Text includes"

• evals.mdx:76 — Accordion title "Text Equals" → "Text equals"

• evals.mdx:84 — Accordion title "Tool Usage" → "Tool usage" ("Tool" stays capped as a product name; "usage" does not)

  Note: If "LLM Judge", "Text Includes", "Text Equals", "Tool Usage" are the exact labels shown in the product UI's type selector, the step instruction text should reference them verbatim (e.g., "Select Text Includes"), but the accordion title itself still follows sentence case per CLAUDE.md.

mcp-server.mdx — product term capitalization

• mcp-server.mdx:7 — "direct access to your agents, tools, and knowledge" → "your Agents, Tools, and Knowledge"
• mcp-server.mdx:241 — "the ability to call Relevance AI tools" → "Relevance AI Tools"; "agents, tools, workforces, knowledge" → "Agents, Tools, Workforces, Knowledge"

🧩 Component suggestions (2)

• evals.mdx:7 — After removing the **Rollout Status**: bold label (see Issues), the <Info> content works fine as a plain sentence. No other component change needed.
• evals.mdx:368–387 — Best practices section uses <CardGroup cols={2}>. CLAUDE.md says CardGroup is not appropriate for "best practices that read naturally as flowing bullets" — these six tips are advisory, not navigable choices. A plain bulleted list or compact two-column prose grid would be more appropriate. If the visual weight of cards is intentional here, it's a mild call; raise with the team if uncertain.

🏗️ Page structure (1)

• evals.mdx:233–235 — The <Note> ends with: "Previously, workforce evaluations always used the active published version regardless of the selection." This is a changelog sentence documenting a past bug, not current behavior. Readers arriving here in six months get noise, not signal. Remove the "Previously…" sentence and keep only the description of what the feature does now.

⚠️ Contradictions (1)

• mcp-server.mdx:247 says "The Relevance AI MCP server exposes the following tools" then lists 4. The neighboring agent-skills.mdx (not changed in this PR) refers to "all 46 MCP tools." The "following" framing implies completeness. If this section is intentionally highlighting only the agent-trigger and evaluation tools (not the full surface area), change the intro to something like: "The Relevance AI MCP server exposes the following tools for triggering agents and running evaluations:" so readers know there are more tools available via the server.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~913

Files read: build/agents/build-your-agent/evals.mdx (429 lines), integrations/mcp/programmatic-gtm/mcp-server.mdx (348 lines), build/agents/build-your-agent/build-overview.mdx (25 lines), integrations/mcp/programmatic-gtm/agent-skills.mdx (111 lines)