arenza-langchain

LangChain.js + LangGraph tools for Arenza — give any agent typed access to AI visibility metrics, GEO opportunities, and brand mention data across ChatGPT, Claude, Gemini, Perplexity, Copilot, and Grok.

Arenza is a Generative Engine Optimization (GEO) platform that measures how 6 leading AI assistants describe brands. This package wraps the Arenza MCP server as LangChain DynamicStructuredTool instances so a LangChain or LangGraph agent can call them like any other tool — with full Zod schemas, GEO-keyword-rich descriptions for solid tool selection, and the same return shapes as @arenza/mcp-client.

Why an Arenza tool plugin (not just a fetch loop)

LangGraph agents that need to answer "how does ChatGPT describe my brand?" or "what should we fix this week to improve our AI search ranking?" benefit from a tool with a tight schema and a clear, GEO-vocabulary description. Hand-rolling fetch calls inside a LangGraph node loses the structured-tool affordance and forces every agent author to re-derive the same parameter shapes.

Install

npm install @arenza/langchain arenza-mcp-client @langchain/core
# or
pnpm add @arenza/langchain arenza-mcp-client @langchain/core

@langchain/core is a peer dependency — install whichever version your LangChain stack already uses (>=0.2.0).

Quick start (LangGraph)

import { ArenzaMCPClient } from '@arenza/mcp-client';
import { getArenzaTools } from '@arenza/langchain';
import { ChatOpenAI } from '@langchain/openai';
import { createReactAgent } from '@langchain/langgraph/prebuilt';

const client = new ArenzaMCPClient({ token: process.env.ARENZA_TOKEN! });

const agent = createReactAgent({
  llm: new ChatOpenAI({ model: 'gpt-4o-mini' }),
  tools: getArenzaTools(client),
  // optional: getArenzaTools(client, { includeWrite: true })  for write tools
});

const res = await agent.invoke({
  messages: [
    { role: 'user', content: 'Which of our brands has the lowest share of voice on Perplexity, and what wrong claims are hurting it?' },
  ],
});

console.log(res.messages.at(-1)?.content);

The agent decides on its own which tools to call. A typical trajectory:

1. arenza_list_brands — to enumerate the portfolio.
2. arenza_get_brand_overview for each — pulling per-LLM mention counts.
3. arenza_list_opportunities on the worst brand, filtered to wrong_claim.

Then the agent stitches a natural-language answer.

Quick start (vanilla LangChain.js, single tool call)

import { ArenzaMCPClient } from '@arenza/mcp-client';
import { getArenzaTools } from '@arenza/langchain';

const client = new ArenzaMCPClient({ token: process.env.ARENZA_TOKEN! });
const [listBrands] = getArenzaTools(client);

const result = await listBrands.invoke({});
console.log(JSON.parse(result));

Tools exposed

getArenzaTools(client) returns 6 read-only tools by default:

| Tool name | What it does | |---|---| | arenza_list_brands | Enumerate brands in the tenant portfolio. | | arenza_get_brand_overview | Share-of-voice + wrong-claim count + per-LLM mentions for one brand. | | arenza_list_prompts | The buyer-perspective prompts being probed for a brand, with mention rate per LLM. | | arenza_list_opportunities | Open GEO opportunities (wrong_claim, missing_canonical_page, listicle_gap, discussion_seed) with severity. | | arenza_suggest_competitors | LLM-suggested competitors to add to tracking. | | arenza_suggest_prompts | LLM-generated buyer-perspective prompts (70%+ unbranded ratio enforced). |

Pass { includeWrite: true } to also get 4 write tools:

| Tool name | What it does | |---|---| | arenza_add_competitor | Add a competitor to a brand's tracking list. | | arenza_dismiss_competitor | Remove a competitor (e.g. wrong suggestion). | | arenza_mark_opportunity_done | Mark a GEO opportunity as completed. | | arenza_generate_geo_article | Draft a canonical-fact article anchored to a specific finding. |

Write tools are opt-in because LangGraph agents can be aggressive about side effects — you usually want a human in the loop before mutating tracking config.

Schemas

Each tool ships a Zod schema, so the underlying LLM (Claude, GPT-4, Gemini) sees the parameter shape and types directly. For example:

// arenza_list_opportunities schema
z.object({
  brand_id: z.string(),
  type: z.enum(['wrong_claim', 'missing_canonical_page', 'listicle_gap', 'discussion_seed']).optional(),
})

The descriptions are deliberately GEO-vocabulary-heavy (mentioning ChatGPT, Claude, Gemini, Perplexity, Copilot, Grok by name; mentioning "share of voice", "hallucinations", "AI visibility") so when an agent's prompt mentions any of those terms the tool router has high signal.

Authentication

Pass an Arenza API token to the client:

const client = new ArenzaMCPClient({ token: process.env.ARENZA_TOKEN! });

Get a token at app.arenza.ai/settings/api. For multi-tenant deployments, swap to OAuth — the MCP server publishes its OAuth metadata at mcp.arenza.ai/.well-known/oauth-authorization-server.

Pattern: weekly GEO triage agent

import { ArenzaMCPClient } from '@arenza/mcp-client';
import { getArenzaTools } from '@arenza/langchain';
import { ChatAnthropic } from '@langchain/anthropic';
import { createReactAgent } from '@langchain/langgraph/prebuilt';

const client = new ArenzaMCPClient({ token: process.env.ARENZA_TOKEN! });
const tools = getArenzaTools(client, { includeWrite: true });

const agent = createReactAgent({
  llm: new ChatAnthropic({ model: 'claude-sonnet-4-5' }),
  tools,
});

const triage = await agent.invoke({
  messages: [
    {
      role: 'system',
      content:
        'You are a GEO marketing triage agent. Each Monday you review the Arenza portfolio. ' +
        'For every brand: list overview, list critical-severity opportunities, draft canonical articles ' +
        'for any wrong-claim opportunity that is older than 7 days, and produce a one-paragraph human summary.',
    },
    {
      role: 'user',
      content: 'Run this week\'s triage.',
    },
  ],
});

Combine with LangGraph persistence to make it stateful across weeks.

Pattern: GEO copilot inside your existing agent

If you already have a customer-facing agent, drop the read-only Arenza tools into the existing tool array — the agent now answers "how am I doing on AI search?" without you wiring custom routes:

const agent = createReactAgent({
  llm,
  tools: [
    ...yourExistingTools,
    ...getArenzaTools(client),  // adds 6 GEO tools
  ],
});

Related projects

• @arenza/mcp-client — the typed TS client this wraps.
• arenza-mcp-client-python — Python equivalent.
• @arenza/cli — npx arenza scan brand.com for terminal scans.
• @arenza/llamaindex — same six tools wrapped for LlamaIndex.
• @arenza/vercel-ai-sdk — Vercel AI SDK provider.
• arenza-zapier-actions — Zapier integration manifest.
• awesome-geo — curated list of GEO and AI-visibility resources.

Resources

• Arenza homepage: https://arenza.ai
• Long-form GEO guides: https://arenza.ai/guides
• AI brand reference: https://arenza.ai/llms.txt + https://arenza.ai/llms-full.txt
• MCP server: https://mcp.arenza.ai
• OAuth spec: https://mcp.arenza.ai/.well-known/oauth-authorization-server
• LangChain docs: https://js.langchain.com

License

MIT (c) 2026 Arenza