The README's own advice is the most useful thing on this page: without the current date, models assume their knowledge cutoff and retrieve stale news. The other habit worth forming is reaching for the stats tools instead of counting search results — sentiment and article volume are computed server-side over the same filters, which is both cheaper and correct. Archiving a monitor cannot be undone through the public API.
The official remote MCP server for Perigon's news API, maintained by Perigon's engineering team. It is large on purpose and lets you shrink it: appending `?tools=` to the URL restricts a session to named tools or a curated profile, which keeps the model focused and the context small.
- Search across articles and stories — `search_news_articles` and `search_news_stories` — plus journalists, sources, people, companies and topics
- Five stats tools that compute aggregates server-side rather than making you count results: `get_avg_sentiment` and `get_article_counts` bucketed over time, `get_top_entities`, and `get_top_people` and `get_top_companies` for coverage spiking against a baseline
- `get_api_access` — this key's scopes, organization, usage quota and the derived behaviour, including blocked filters and date-window clamps; it does not count against quota
- Monitors — `list_monitors`, `get_monitor`, `get_monitor_events`, `get_monitor_newsletters`, `get_monitor_summaries`, `set_monitor_status`, with `create_monitor` and `update_monitor` available only when you opt in
- Platform tools — `get_source_by_id`, `get_top_topics`, `get_story_stats`, `watchlists`, `source_groups`, `contact_points`, `article_refresh`, and create/update variants for watchlists and source groups
- Signal Insights, on keys with that scope — `signal_insights_create_workspace` once per conversation, then search, read and export tools plus a persistent Python sandbox with `signal_insights_execute_code`, `signal_insights_preview_chart`, `signal_insights_shell` and file tools; pandas, numpy and matplotlib come pre-installed
- Six reusable prompts that encode multi-step chaining — `entity_deep_dive`, `narrative_trace`, `coverage_trend`, `journalist_beat_profile`, `competitive_landscape`, `spike_explainer`
- Reference material as on-demand resources: `perigon://reference/fields`, `perigon://reference/chaining`, `perigon://reference/entitlements`, `perigon://reference/charts`
A Perigon API key, sent as `Authorization: Bearer`. Nothing to install: add `https://mcp.perigon.io/v1/mcp` as a Streamable HTTP server, or reach it through `npx mcp-remote@latest` if your client has no native HTTP support. The SSE endpoint at `/v1/sse` exists for legacy clients and is not recommended for new setups. Which tools you see depends on your key's scopes — the `?tools=` parameter can only narrow that set, never widen it. Profiles are `research`, `monitoring`, `platform` and `minimal`, and can be combined with individual names.
One command plus a key — claude mcp add --transport http perigon_news_api https://mcp.perigon.io/v1/mcp \ --header "Authorization: Bearer YOUR_PERIGON_API_KEY", then supply credentials