MentionFox MCP documentation

Everything you need to connect MentionFox to Claude, Cursor, ChatGPT, Windsurf, Continue.dev, or any AI agent that speaks Model Context Protocol. Setup, tool reference with example prompts, authentication, troubleshooting, and privacy.

Overview

Model Context Protocol (MCP) is an open standard published by Anthropic in late 2024 for connecting AI agents to external services. The agent gets a list of tools it can call; when it decides to use one, it sends a structured JSON-RPC request, the service runs server-side, and the result lands back in the conversation as part of the agent's context.

MentionFox exposes 9 of its endpoints over MCP at https://www.mentionfox.com/mcp. Any MCP-speaking client connects in under 60 seconds. Once connected, your AI agent can research people, vet investors and founders, find journalists, scan platforms for mentions, enrich contacts, save leads to your pipeline, and recall what you've already researched — all without you context-switching out of the agent's chat surface.

This page is the deeper companion to the landing page. If you only want to connect, the 60-second setup over there is enough. This page is for when you want to know exactly what each tool does, how the credit pricing works, what the auth flow looks like, and how to troubleshoot when something behaves oddly.

Why use MentionFox via MCP

The MentionFox dashboard is a great place to do focused research. The MCP surface is for everything else — the moments when you're already deep in a Claude or Cursor or ChatGPT conversation and you need a piece of intel without breaking your flow. Three reasons users connect:

Stay in your tool of choice. Switching to the dashboard, searching, copying the result, pasting back into Claude is friction. With MCP the result lands directly in the conversation. The agent reads it, you read it, the conversation continues.

Let the agent chain the work. "Research Sara at Acme, then enrich her, then save her to my pipeline as a Q3 prospect" — three tool calls, one prompt. The agent orchestrates; you don't.

Better use of structured output. Humans skim UI cards. Agents parse the full structured payload — the bio, the recent activity, the contact hints, the data-quality score — and reason about it more thoroughly than a human glance ever could.

If you've never used MCP before, the easiest first run is: connect to Claude, type "Tell me about any name you know" and watch what happens. The 60-second feedback loop sells the rest.

Setup per client

Claude (web + desktop)

1. Open Claude → Settings → Connectors
2. Click Add Connector
3. Enter https://www.mentionfox.com/mcp as the server URL
4. Authorise via the OAuth popup — sign in to MentionFox, approve scope
5. Done. Tools appear in any Claude conversation immediately.

If your Claude session is the desktop app, the OAuth popup opens in your default browser. Same flow.

Cursor IDE

1. Open Cursor → Settings → MCP tab
2. Click + Add new MCP server
3. Paste the JSON config (below)
4. Save → Cursor will open a browser tab for OAuth. Sign in, approve.

{
  "mcpServers": {
    "mentionfox": {
      "url": "https://www.mentionfox.com/mcp"
    }
  }
}

Tools then surface in Cursor agent chat. They're available to any Cursor agent, including @-mentioned tools in inline edits.

Windsurf

1. Open Windsurf → Settings → Cascade → MCP servers
2. Click Add MCP server
3. Use the same JSON config as Cursor (above)
4. Authorise via the OAuth flow

Continue.dev

Add to your ~/.continue/config.json:

{
  "mcpServers": [
    {
      "name": "mentionfox",
      "transport": "http",
      "url": "https://www.mentionfox.com/mcp"
    }
  ]
}

Reload Continue. The first tool call will redirect to the MentionFox OAuth flow.

ChatGPT custom GPTs (via MCP gateway)

ChatGPT doesn't speak MCP natively yet, but you can bridge via an MCP-to-Actions gateway. The MentionFox connector exposes a standard OpenAPI shim at https://www.mentionfox.com/connectopenapi.json for this case. Import the OpenAPI URL into your custom GPT's Actions config and use the API key authentication path (described under Authentication below).

Anthropic SDK / OpenAI SDK direct integrations

Both SDKs ship MCP client primitives. Add MentionFox as a tool source:

// Anthropic SDK (TypeScript)
import { Anthropic } from "@anthropic-ai/sdk";
const client = new Anthropic();
const response = await client.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  mcp_servers: [
    { type: "url", url: "https://www.mentionfox.com/mcp", name: "mentionfox" }
  ],
  messages: [{ role: "user", content: "Research Sara Chen at Acme Corp" }]
});

Generic agent frameworks (LangChain, CrewAI, AutoGen, custom)

Any framework with MCP support: register the URL https://www.mentionfox.com/mcp, configure OAuth or API-key auth (next section), tools surface automatically. If your framework doesn't yet speak MCP, the OpenAPI shim above gives you a vanilla REST surface with the same 9 endpoints.

Authentication

OAuth (recommended for interactive clients)

The MCP endpoint runs a standards-compliant OAuth 2.0 authorisation server with the following endpoints:

• POST /functions/v1/mcp-oauth-register — dynamic client registration (RFC 7591)
• GET /functions/v1/mcp-oauth-authorize — authorisation request
• POST /functions/v1/mcp-oauth-token — token exchange + refresh
• GET /functions/v1/mcp-oauth-userinfo — userinfo

Most MCP clients (Claude, Cursor, Windsurf) handle this transparently. You see one OAuth popup, sign in, and that's it. Tokens are scoped to your MentionFox user and refresh automatically.

API key (recommended for non-interactive clients and server agents)

For agents running server-side (cron, webhooks, custom backends), use an API key:

1. Open MentionFox dashboard → Settings → API keys
2. Click Create MCP key
3. Copy the mfx_... token (shown once)
4. Pass it as Authorization: Bearer mfx_... on every MCP request

API keys never expire automatically; revoke from the same page when needed. They count against the same monthly cap as OAuth-authenticated calls.

9 tools reference

Each tool is documented with its full input schema, what it returns, the credit cost, and three example prompts you can paste directly into your AI client to test it.

Pricing + credit costs

One subscription, one credit pool, three plans. Trial is free for 7 days with 500 calls included; you don't even need to add a card.

Per-tool credit costs (full table):

The Tier 1 pool is a shared monthly call counter for the four "deep research" tools — research_person, research_company, get_investor_report, get_entrepreneur_report. Each call to any of those four counts as one against the pool. Once the pool is empty for the month, those four tools hard-block until the next month or until you upgrade.

The enrich_person cap is separate — it has its own monthly call limit and doesn't pull from the Tier 1 pool. Same enforcement: hard-block when reached.

The find_journalists_covering and scan_for_mentions tools are uncapped within your credit budget. Use them as much as your credit pool can fund.

Above the per-tier caps, contact hi@mentionfox.com for a custom contract — typically priced for agencies and large API consumers.

Troubleshooting

Top 10 issues seen in the wild, in rough order of frequency:

1. "Quota exceeded" error on a Tier 1 tool

You've hit the per-month Tier 1 pool cap (10 / 50 / 500 calls per Starter / Growth / Scale). Upgrade your plan or wait until the next month. The pool resets at the start of every calendar month UTC.

2. OAuth popup never appears

Browser pop-up blocker. Enable popups for the AI client's domain (claude.ai, cursor.sh, etc.) and retry. The OAuth flow opens a fresh tab; if blocked, it never gets to the MentionFox login page.

3. Tool returns "Insufficient public data"

The research engine couldn't find enough public signal to give you a high-confidence result. This is intentional — MentionFox would rather return nothing than guess. Try adding a company hint or LinkedIn URL to the call to disambiguate. These calls are not charged.

4. Tool times out (25 second wall)

The MCP gateway has a 25-second per-call wall to prevent agents from hanging. Some deep enrichments take longer; the gateway returns a "queued — open dashboard for the full result" pointer in that case. The dashboard view typically completes in 60-90 seconds.

5. "Unknown tool" error

Your MCP client cached an old tools/list. Disconnect and reconnect the connector to refresh. Most clients also support a "Reload tools" action in the connector settings.

6. Different tool list than expected

The current launch surface is 9 tools. If you see a different number, your client may have stale metadata or you may be hitting a beta endpoint. Verify by curling https://www.mentionfox.com/mcp with a JSON-RPC tools/list call.

7. API key returns 401

Either the key was revoked, the key was never created, or you're using the wrong header name. Format: Authorization: Bearer mfx_<32-char-token>. The leading "mfx_" is part of the token, not a prefix to strip.

8. Tool calls show in dashboard but agent says "no results"

Your AI agent is choosing to ignore the tool output and answer from its own knowledge. This is rare on Claude but happens with some other agents. Try a more explicit prompt: "Use the research_person tool to look up Sara Chen, then summarise the response."

9. Subscription expired mid-call

The call returns 402 with a clear message. Your saved data (pipeline, dossiers, recent research) is preserved for 90 days; reactivating your subscription restores access to all of it instantly.

10. Multi-account confusion

If you have multiple MentionFox accounts (e.g. personal + work), each requires its own MCP connection. Disconnect, sign in to the desired account in your browser, then reconnect — the OAuth popup will pick up the active session.

Privacy

The MCP-specific privacy posture supplements the broader MentionFox privacy policy:

Data sources are public record. MentionFox does not scrape behind logins, does not buy data from broker networks, does not use private records. Sources are public web — published bylines, public profiles, official websites, news articles, public registries.

Your tool inputs are logged for billing and tier-cap enforcement. We log: tool name, timestamp, success / failure, credits charged, plus the arguments object (so you can audit your own usage at Settings → MCP Activity). Logs are retained for 12 months.

Tool inputs are NOT used to train any model. No part of your MCP traffic feeds any LLM training set, ours or anyone else's. The research engine itself uses standard third-party LLM APIs (Anthropic, OpenAI) to synthesise findings — those calls fall under each provider's own data-handling policy, which all confirm no-training-on-API-traffic by default.

Cancellation. Your account remains usable through the end of the billing period. Saved data persists for 90 days post-cancellation. After that, hard-deleted unless you've reactivated.

Data subject rights (GDPR / CCPA / similar). Email privacy@mentionfox.com for export, deletion, or correction requests. We respond within 30 days.

Audit trail. Every MCP call lands in the dashboard's Settings → MCP Activity view — full transparency on what your AI agents called, when, and what they were charged. Build trust between you and your agents by auditing this trail occasionally.

Documentation last updated 2026-05-02. For the launch landing page, see /mcp. Questions: hi@mentionfox.com.