Heron Intelligence MCP

Connect Heron Intelligence — curated earnings call transcripts, management interviews, and a guided custom-research intake flow — directly to Claude or any Model Context Protocol-compatible AI client.

Highlights

• Remote HTTPS transport — one URL, zero install for browser clients.
• OAuth 2.1 — Dynamic Client Registration (RFC 7591) with PKCE S256.
• Subsidiary & legacy-name resolution — search by ticker, name, or description; the server auto-maps spinoffs and rebrands to their current parent so coverage is never missed.
• MCP tool annotations — read-only tools auto-approve in supporting clients; write tools always require explicit user confirmation.

Server

https://mcp.heron-intelligence.com/mcp

Skill (recommended — install first)

The repo ships a Claude Code Skill in skills/heron-intelligence/ that primes the model to use these tools automatically — without the user having to type "use Heron MCP to…". It also encodes formatting rules and the canonical provenance handling: Blue Heron (human-conducted), Heron AI (AI-conducted), and licensed third-party feeds such as Paragon — each named explicitly so the origin of every transcript stays visible to the user.

Install once per machine (org-wide):

cp -r skills/heron-intelligence ~/.claude/skills/

Restart Claude Code; verify with /list-skills. See skills/heron-intelligence/INSTALL.md for project-level install and team distribution.

Quickstart

Claude.ai (recommended)

1. Open Claude.ai → Settings → Connectors.
2. Click Add custom connector.
3. Enter the server URL above.
4. Click Authorize and sign in to Heron Intelligence.
5. Approve the requested scopes (mcp:read, mcp:write).
6. Start a new chat and ask Claude to use Heron Intelligence.

Claude Desktop

Add to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "heron": {
      "command": "uvx",
      "args": ["--from", "HeronIntelligence-mcp", "heronintelligence-mcp"],
      "env": { "HERON_API_KEY": "hk-xxxx" }
    }
  }
}

Restart Claude Desktop.

Claude Code

claude mcp add heron \
  --command uvx \
  --args "--from HeronIntelligence-mcp heronintelligence-mcp" \
  -e HERON_API_KEY=hk-xxxx \
  -s user

Cursor / Windsurf

{
  "mcpServers": {
    "heron": {
      "command": "uvx",
      "args": ["--from", "HeronIntelligence-mcp", "heronintelligence-mcp"],
      "env": { "HERON_API_KEY": "hk-xxxx" }
    }
  }
}

Authentication

The remote MCP server uses OAuth 2.1 with PKCE (S256) and Dynamic Client Registration (RFC 7591). Connecting from Claude.ai:

1. Claude registers itself silently as an OAuth client.
2. You land on a Heron Intelligence consent page that names the application requesting access and asks for the Client ID and Client Secret issued to your seat. Get them from your account dashboard, or from your Heron administrator.
3. After consent, Claude receives a short-lived access token plus a refresh token. The refresh token is single-use and rotates on every refresh.
4. The server advertises the scopes mcp:read and mcp:write, and Claude requests both. Access to a given tool is decided by your own entitlements on the Heron Intelligence account behind the token.

For local stdio mode (Claude Desktop / Claude Code / Cursor / Windsurf), authenticate with a static API key via HERON_API_KEY. Request a key from your Heron Intelligence account dashboard.

Available tools

Read-only — research

Read-only — watchlists

Read-only — alerts (saved monitoring rules)

Write — custom research projects

Write — watchlists

Write — alerts

The alert tools were called drops before the Portal renamed the module. Only the alert names are listed in tools/list; a call that still uses an old name (list_drops, create_drop, ...) is served by the alert tool and its response carries renamed_to with the name to use.

Access by plan

Heron Events Platinum firms (firm_type = he_premium) do not get the Research Projects or Evasion tools: they are left out of tools/list and a call by name returns {"error": "tool_not_in_plan", ...}. The caller's tier comes from the API's GET /api/v1/auth/caller; when it cannot be read, nothing is restricted.

Call date vs post date

Every transcript has two dates, and they can be months apart: due-diligence interviews are often posted well after the call.

get_recent_transcripts options:

• sort_by (call_date or post_date) and sort_order (desc or asc). When no range is given, the quarters window applies to the sort_by date. So sort_by="post_date", quarters=2 returns what was posted in the last six months, newest first.
• Explicit ranges (YYYY-MM-DD, inclusive, either end optional). call_date_from/call_date_to and post_date_from/post_date_to replace the quarters window. Ranges on both dates combine with AND. An empty result is reported as such (coverage.date_range_applied: true) and the range is never widened.
• applied_constraints echoes window_field, sort_by, sort_order and any ranges, so the answer can state which date it is describing.

sort_by and sort_order order the browsing path. A free-text query is ranked by relevance and ignores them.

Example prompts

"How much research coverage does Heron Intelligence have on Amazon?"

"Find Heron Intelligence's research coverage for Facebook and show me what's available under their current corporate entity."

"Show me only the management due diligence transcripts available for Amazon from the last two quarters."

"What Heron Intelligence transcripts on Netflix were posted in the last two months, newest first?"

"Set up a weekly digest of new Energy-sector transcripts that mention margin compression."

"What alerts do I have set up? Show me the ones that fired this week."

"Add MSFT and NVDA to my Tech watchlist, then create an analysis alert on that watchlist using the Executive Assessment category."

Documentation

Full documentation: https://docs.heron-intelligence.com/overview/mcp

Privacy Policy

https://www.heron-intelligence.com/privacy-policy

The policy covers what is collected, how it is used and stored, third-party sharing, retention, and who to contact. What this MCP server itself handles:

• Identity. OAuth resolves the connecting user to their Heron Intelligence account; whoami returns that email. No other profile data is read.
• Tool arguments and results. Queries are forwarded to Heron Intelligence's own first-party API and the response is returned to the client. Nothing is proxied to a third party.
• Operational records. One structured access line per request (method, path, status, latency, request id, and the resolved account), with the credential redacted out of the path before the line is written. Each tool call also records a usage event to Heron Intelligence's own backend — tool name, the arguments it was called with, a short summary of the result, status and duration — used for quota enforcement and support.
• Not collected. Conversation content, chat history, memory, or user files. The server reads only the arguments a tool is called with.

Support

• Email: support@heron-intelligence.com
• Issues: https://github.com/Heron-Intelligence/heron-intelligence-mcp/issues

License

Released under the Apache License 2.0 — see LICENSE. Use of the hosted Heron Intelligence service requires an active account.