docs(TSP-1306): document new MCP tools for large agent system prompts

[claude]

Summary

• Adds a new Available MCP tools section to the MCP Server page (/integrations/mcp/programmatic-gtm/mcp-server) documenting three new tools that enable AI assistants to work with large agent system prompts:
  • relevance_read_agent_system_prompt — paginated, line-numbered window reading (auto-approved)
  • relevance_search_agent_system_prompt — regex search with context lines (auto-approved)
  • relevance_edit_agent_system_prompt — targeted string replacement saved as draft (requires user approval)
• Notes the new system_prompt_line_count field on relevance_get_agent
• Adds a brief tip on the Programmatic GTM overview page pointing to the new section for large prompt use cases

Context

Implements documentation for PR #15594 (new MCP tools for large system prompts).

Linear: https://linear.app/relevance/issue/TSP-1306/

Test plan

• Verify the new section renders correctly in Mintlify preview
• Check all anchor links resolve (#available-mcp-tools)
• Confirm accordion titles are sentence case
• Confirm no banned words (powerful, seamlessly, etc.)

🤖 Generated with Claude Code

[linear]

TSP-1306

[mintlify]

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

Project	Status	Preview	Updated (UTC)
relevanceai	🟢 Ready	View Preview	Jun 17, 2026, 3:49 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	8/10	Two British spellings ("behaviour") in programmatic-gtm.mdx. No banned words, capitalization is clean throughout.
🟡	Technical clarity	8/10	MCP tool parameter documentation in mcp-server.mdx uses dense prose instead of tables — harder to scan for specific parameter names, defaults, and descriptions. Everything else is specific and accurate.
🟢	Non-technical clarity	9/10	No issues found. Definitions are clear before instructions begin, and the example prompts section is genuinely useful.
🟡	Structure	7/10	programmatic-gtm.mdx is a concept/landing page with no closing CTA. Bold text used as section labels inside Tabs instead of proper headings.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5. Scores cover the whole PR.

✨ Overall vibe: Both pages are well-written with accurate, specific content and good component usage throughout. The MCP tools reference section in mcp-server.mdx is thorough and the example prompts on the concept page do real explanatory work. Main things to fix: two stray British spellings, a missing CTA on the landing page, and parameter prose that would scan better as tables.

🔧 Issues (2)

• get-started/core-concepts/programmatic-gtm.mdx:54 — behaviour → behavior (British spelling in the "Refine agents" card description)
• get-started/core-concepts/programmatic-gtm.mdx:135 — behaviour → behavior (British spelling in the "Troubleshoot" tab intro)

🧩 Component suggestions (2)

• integrations/mcp/programmatic-gtm/mcp-server.mdx:253–256, 267–270, 281–285 — Each tool's parameters are described in flowing prose (agent_id identifies which agent… start_line sets the first line… num_lines controls…). These are reference parameters with consistent attributes — convert each block to a small markdown table with columns Parameter, Default, and Description. This applies to all three tool descriptions (relevance_read_agent_system_prompt, relevance_search_agent_system_prompt, relevance_edit_agent_system_prompt).

• get-started/core-concepts/programmatic-gtm.mdx:74, 95, 116, 136 — **Example prompts:** is bold text acting as a sub-section label inside four <Tab> components. CLAUDE.md says not to mix bold text with heading formats for section titles. Replace with #### Example prompts (no trailing colon).

🏗️ Page structure (1)

• get-started/core-concepts/programmatic-gtm.mdx — No closing CTA. This is a concept/landing page; it should direct readers somewhere after the FAQ section. Natural next steps that exist in the repo: /integrations/mcp/programmatic-gtm/claude-code (quickest path to get started), /integrations/mcp/programmatic-gtm/mcp-server (for non-Claude-Code clients), /integrations/mcp/programmatic-gtm/agent-skills (to give the AI assistant Relevance AI knowledge). A ## What's next? with two or three cards would round the page off.

✅ Clean files (0)

No fully clean files in this PR.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~859

Files read: get-started/core-concepts/programmatic-gtm.mdx (219 lines), integrations/mcp/programmatic-gtm/mcp-server.mdx (403 lines), integrations/mcp/programmatic-gtm/agent-skills.mdx (111 lines), integrations/mcp/programmatic-gtm/claude-code.mdx (126 lines)

[claude]

Corrected parameter names based on product PR #15594: start_line→offset_line, num_lines→limit_lines, context_lines→context, old_text→old_string, new_text→new_string. Also added documentation for the limit_lines: 0 behavior (returns size only).

[github-actions]

🎯 Vibe check

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

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	Five British spellings in programmatic-gtm.mdx (behaviour ×2, personalised ×2, categorises ×1); product feature terms (Agent, Tool, Workforce) consistently lowercased throughout both files when clearly referring to the Relevance AI product.
🟢	Technical clarity	9/10	Near-solid. One time-sensitive phrase in mcp-server.mdx ("now includes") will age poorly.
🟡	Non-technical clarity	7/10	"GTM" is used throughout programmatic-gtm.mdx without ever being expanded to "Go-to-Market". Readers from outside a sales/marketing background will not know what it means.
🟢	Structure	9/10	Both pages are logically organised. Minor: programmatic-gtm.mdx is a concept/hub page but ends on an FAQ with no closing CTA.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: Both pages are well-written, well-structured, and cover their subject clearly — the MCP server page in particular is detailed and production-ready. The main drag is consistency: five British spellings and a pattern of lowercased product feature terms (Agent, Tool, Workforce) across both files that will need a pass before merge.

🔧 Issues (10)

British spellings — programmatic-gtm.mdx

• get-started/core-concepts/programmatic-gtm.mdx:54 — behaviour → behavior (inside Card description: "Iterate on agent instructions, tool configurations, and behaviour based on real results.")
• get-started/core-concepts/programmatic-gtm.mdx:82 — personalised → personalized (inside example prompt: "send a personalised follow-up email via Gmail")
• get-started/core-concepts/programmatic-gtm.mdx:84 — categorises → categorizes (inside example prompt: "categorises messages by urgency")
• get-started/core-concepts/programmatic-gtm.mdx:129 — personalised → personalized (inside example prompt: "generates a personalised cold email")
• get-started/core-concepts/programmatic-gtm.mdx:135 — behaviour → behavior (inside Tab content: "dig into agent behaviour, tool failures, and configuration problems")

Product term capitalization

CLAUDE.md requires Agent, Tool, Workforce, Trigger, Knowledge to be capitalized when referring to Relevance AI product features. Both files use lowercase throughout, including in contexts where the product feature is clearly the referent.

• get-started/core-concepts/programmatic-gtm.mdx:35–62 — Card titles "Create agents", "Build tools", "Set up workforces", "Trigger agents", "Execute tools", "Troubleshoot agents", "Refine agents" → "Create Agents", "Build Tools", "Set up Workforces", "Trigger Agents", "Execute Tools", "Troubleshoot Agents", "Refine Agents"
• get-started/core-concepts/programmatic-gtm.mdx:71,113 — Tab titles "Build agents" → "Build Agents", "Build tools" → "Build Tools"
• integrations/mcp/programmatic-gtm/mcp-server.mdx:241 — "your agents, tools, and workforces" → "your Agents, Tools, and Workforces"
• integrations/mcp/programmatic-gtm/mcp-server.mdx:309 — "your agents, tools, and workforces" → "your Agents, Tools, and Workforces"

Time-sensitive language

• integrations/mcp/programmatic-gtm/mcp-server.mdx:249 — "The relevance_get_agent tool now includes a system_prompt_line_count field" — "now" reads as a release note and will become confusing once this is no longer new. Drop it: "The relevance_get_agent tool includes a system_prompt_line_count field…"

🏗️ Page structure (2)

• get-started/core-concepts/programmatic-gtm.mdx — No closing CTA. This is a concept/hub page: readers who make it to the end of the FAQ are engaged but have no next step. Add a ## What's next? or closing <CardGroup> pointing to the setup paths — e.g. /integrations/mcp/programmatic-gtm/claude-code, /integrations/mcp/programmatic-gtm/mcp-server, /integrations/mcp/programmatic-gtm/agent-skills.

• get-started/core-concepts/programmatic-gtm.mdx:7 — "GTM" is used in the opening sentence ("Programmatic GTM is the new way to build your agents for GTM in Relevance AI") without ever being expanded. Expand on first use: "Programmatic GTM (Go-to-Market)". Readers who arrive at this page from search or a direct link may not know the abbreviation.

✅ Clean files (0)

Both files have at least one issue.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~935

Files read: get-started/core-concepts/programmatic-gtm.mdx (219 lines), integrations/mcp/programmatic-gtm/mcp-server.mdx (403 lines), get-started/core-concepts/agents.mdx (202 lines), integrations/mcp/programmatic-gtm/agent-skills.mdx (111 lines)