Runnable, forkable example agents that call the Meltwater MCP server to do real PR & comms work — crisis monitoring, executive briefings, campaign insight — as code that runs in your systems. Not a chatbot and not a dashboard: agents your team owns, schedules, and wires into its own data.
Each example is a complete starting point, not tested production code. Fork one and you can have a working proof-of-concept immediately. Tweak, extend, polish and test the code to make it yours.
🗞️ Exec Media Brief — meltwater-mcp/exec-media-brief
A branded, executive-ready media brief — news + social, selected, ranked and rendered to a self-contained HTML page — delivered headless every morning, before anyone's awake. Pure Meltwater, one credential, a great-looking artifact on day one.
🚨 Reputation Watchdog — meltwater-mcp/reputation-watchdog
Scans a brand's coverage for an anomaly against its 30-day norm, then fuses it with your own first-party data (support desk, CRM, claims) into an alert neither source could give alone — "the same regions were lighting up in our own data days before the story broke." Shows how to join Meltwater with a private internal system, inside your own agent.
🧭 Social Brief Sub-Agent — meltwater-mcp/social-brief-subagent
Already building your own agent platform? Plug Meltwater in as a specialist insight sub-agent: your orchestrator asks for a grounded landscape read on a brand × market × topic, and Meltwater returns the trending stories, cultural moments and whitespace — your agent writes the campaign brief. Meltwater as a source of insight inside your agentic stack.
cd meltwater-mcp/exec-media-brief
uv sync # installs deps
cp .env.example .env # add your Anthropic API key + Meltwater MCP token
uv run python exec_media_brief.py --out brief.html
open brief.html # a finished, branded brief — prints straight to PDFEvery example follows the same shape and ships with its own README, a fictional demo config you
can run as-is, and a no-code path (below). Prefer to look before you install? Each folder's
sample_output/ shows a finished result.
- An Anthropic API key — the agent's reasoning (console.anthropic.com → API Keys).
- A Meltwater MCP API token — access to your Meltwater data (from your Meltwater account team).
Drop them in .env and go — nothing is hardcoded, so nothing leaks when you share your fork.
- No code (fastest look) — let an MCP client like Claude Code act as the agent via your
connected Meltwater connector. See each example's
RUNNING-IN-CLAUDE-CODE.md. - Standalone Python — the production-shaped path; runs entirely in your own process, one Meltwater credential as the only external seam.
- Headless / scheduled — drop it on cron / Lambda / Cloud Run so it runs itself and lands in your channel. That self-running loop is the reason to build an agent rather than do it by hand.
The value is the orchestration you own: it runs on your schedule, in your infrastructure, joined with your own data, delivered to your channel — no human in a chat. These examples teach that pattern on real PR & comms workflows, as code you can fork and ship.
The Model Context Protocol is an open standard for giving AI agents tools. Meltwater MCP exposes Meltwater's news & social intelligence as MCP tools, so your agent can query coverage, sentiment, sources and posts directly. For the full API reference, tool catalogue and quickstart, see the Meltwater Developer Portal at developer.meltwater.com — this repository provides the runnable examples that complement it.
Examples are grouped by the Meltwater product they use, then by use case:
meltwater-mcp/
exec-media-brief/ # scheduled, branded daily media brief (HTML)
reputation-watchdog/ # anomaly detection fused with your own first-party data
social-brief-subagent/ # Meltwater as an insight sub-agent to your orchestrator
Draft reference implementations — starting points to fork, not turnkey products. Confirm SDK calls against your installed versions before production use.