MCP Tiers

The edgar.tools MCP server advertises a tier-aware tool list. When your AI client calls tools/list, it sees the tools your account's tier includes — not the full surface. The prompts/list endpoint behaves the same way for recipe prompts. Cross-tier invocations return a structured tier_denied envelope so the AI can route around it gracefully.

This page is the canonical mapping. The visual ladder lives at the AI plugin landing page; this page is the operational reference for engineers and analysts who need to know exactly what each tier exposes.

Source of truth. The mapping below mirrors src/lib/mcp/tier-map.ts in the open server. If a tool ships at a different tier than this page says, the server wins and the page is stale — file an issue.

Quick reference#

Tools and prompts accumulate as you climb tiers — Pro sees Free's tools plus Pro's, Analyst sees everything below.

Free — Read#

The starter surface. Every account gets this; no payment required.

Free tier returns truncated results on these tools (e.g. 10 filings per call, basic profile data). Upgrade prompts surface in the response when fuller data is available at a higher tier.

Pro — Track#

Pro adds the research toolkit on top of Free. The shape: financial intelligence, signals, lookup feeds, and the fund/adviser surface.

Plus everything Free includes.

Analyst — Research#

Analyst adds the analysis and aggregation layer, plus the recipe prompts, on top of Pro. The shape: AI-powered filing analysis, person- and account-level intelligence, and an aggregation layer designed to be consumed by recipe prompts.

Plus everything Pro includes.

Analyst also unlocks five recipe prompts available via prompts/list and prompts/get. In Claude Desktop and Claude.ai they appear under the edgar slash-command namespace.

See MCP Prompts for the per-recipe argument tables and example invocations.

How tier gating works#

Two layers, so a client can't bypass it:

1. `tools/list` filter. The MCP server inspects your bearer token's tier and returns only the tools your tier includes. The prompts/list endpoint applies the same filter — recipe prompts only appear at Analyst+. The AI never knows tools or prompts above your tier exist.
2. Invoke-time check. Even if a client somehow constructs a call to a higher-tier tool (cached schemas from a prior session, manual JSON-RPC), the dispatcher rejects with a structured tier_denied error envelope:

{
  "error": {
    "code": "tier_denied",
    "message": "This tool requires the Analyst tier.",
    "required_tier": "analyst",
    "current_tier": "pro",
    "upgrade_url": "https://app.edgar.tools/pricing"
  }
}

The AI assistant reads tier_denied and routes around it — typically by surfacing the upgrade URL to you in chat or by trying a lower-tier alternative.

After upgrading#

When you upgrade tiers, your existing token still has the old tier baked in. To pick up the new tools:

1. Disconnect the edgar.tools server in your MCP client (Claude → Settings → Connectors → ⋯ → Disconnect, or remove from mcp.json).
2. Reconnect using the same URL. OAuth issues a fresh token tagged with your new tier.
3. The tool list refreshes — the newly unlocked tools appear, and at Analyst+ the recipe prompts appear in prompts/list.

Tokens otherwise last 90 days. After a tier downgrade or cancellation, the next refresh issues a token at the new tier.

What's next#

• MCP landing page — visual tier ladder + setup walkthroughs for Claude, ChatGPT, Cursor, Windsurf
• MCP Setup — connect your AI client step by step
• MCP Tools reference — parameter tables and per-tool behaviour
• MCP Prompts reference — the 5 Analyst recipe prompts
• Pricing — full tier comparison including web-app features
• Dataset catalog — need bulk data instead of live tool calls? Per-dataset subscriptions from $24.99/mo, or a bespoke annual license for commercial use