Perigon MCP Server

Perigon MCP Server

This is the official MCP server for the Perigon news API.

Documentation

For more information on how to use and connect the MCP, visit the MCP docs.

Maintainers

The Perigon MCP Server is developed and maintained by the Perigon engineering team.

Lead Developer: Islem Maboud, responsible for the architecture, implementation, and ongoing development of this MCP server, including the remote transport layer, authentication handling, deployment configuration, and the MCP playground.

Usage

Playground

You can try out the Perigon MCP server in our playground.

Note: A valid Perigon API key is required to use the MCP. The MCP playground requires you to be already authenticated to the Perigon dashboard.

Connecting

You can connect to our remote MCP server using any MCP-compatible client.

Server URL: https://mcp.perigon.io

The recommended transport is Streamable HTTP (/v1/mcp). SSE (/v1/sse) is supported for legacy clients but not recommended for new integrations.

Quick Setup Examples

Streamable HTTP — native support (recommended):

{
  "mcpServers": {
    "perigon_news_api": {
      "url": "https://mcp.perigon.io/v1/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer YOUR_PERIGON_API_KEY"
      }
    }
  }
}

Streamable HTTP — via mcp-remote (for clients without native HTTP support):

{
  "mcpServers": {
    "perigon_news_api": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@latest",
        "https://mcp.perigon.io/v1/mcp",
        "--header",
        "Authorization: Bearer ${PERIGON_API_KEY}"
      ],
      "env": {
        "PERIGON_API_KEY": "YOUR_PERIGON_API_KEY"
      }
    }
  }
}

For Claude Code (CLI):

claude mcp add --transport http perigon_news_api https://mcp.perigon.io/v1/mcp \
  --header "Authorization: Bearer YOUR_PERIGON_API_KEY"

SSE (legacy clients only):

{
  "mcpServers": {
    "perigon_news_api": {
      "url": "https://mcp.perigon.io/v1/sse",
      "type": "sse",
      "headers": {
        "Authorization": "Bearer YOUR_PERIGON_API_KEY"
      }
    }
  }
}

📖 For detailed setup instructions for different clients, see our comprehensive MCP documentation.

Selecting specific tools

By default all tools permitted by your API key are available. You can restrict a session to a smaller set by appending a ?tools= query parameter to the server URL. This is useful for reducing context size and keeping the model focused.

https://mcp.perigon.io/v1/mcp?tools=search_news_articles,search_news_stories

• Pass a comma-separated list of tool names, or all to explicitly activate every permitted tool.
• Only tools your API key already has access to will be activated — the parameter cannot expand permissions.
• Omitting the parameter, passing an empty value, or passing all are all equivalent and activate every permitted tool.

Example — Cursor config scoped to article and story search:

{
  "mcpServers": {
    "perigon_news_api": {
      "url": "https://mcp.perigon.io/v1/mcp?tools=search_news_articles,search_news_stories",
      "type": "http",
      "headers": {
        "Authorization": "Bearer YOUR_PERIGON_API_KEY"
      }
    }
  }
}

Named tool profiles

Instead of listing individual tool names, ?tools= also accepts curated profile aliases — useful shorthand for reducing context to a task-appropriate subset:

Profile	Includes
research	Core search, stats, and lookup tools for news research (articles, stories, journalists, sources, people, companies, topics, stats, get_source_by_id, get_api_access).
monitoring	All monitor tools (including create_monitor/update_monitor) plus every Signal Insights tool.
platform	Watchlists, source groups, contact points, article refresh status, and get_api_access.
minimal	search_news_articles plus the five stats tools and get_api_access — the smallest useful research set.

Profiles can be combined with explicit tool names in the same ?tools= value (e.g. ?tools=research,create_monitor), and are always intersected with what your API key's scopes actually permit.

Prompt Examples

When prompting your agent we recommend providing the current date (or a tool to get it) unless the agent already has access to such information, this is because some models like Claude will otherwise think the current date is their knowledge cutoff and they will retrieve outdated information frequently.

News Articles & Stories:

• Give me the top 5 political headlines in the United States from today.
• What business stories are trending in New York today?
• Show me the latest tech news from California this week.
• Find political news from swing states in the last 3 days.
• Show me cryptocurrency-related stories from the past week.

Journalists & Sources:

• Find local news sources in Texas.
• Who are the top business journalists at major publications?
• Find journalists covering renewable energy, then show me their recent articles.
• Which journalists write the most about climate policy?
• Show me articles from major financial publications today.

People & Companies:

• Find recent news about pharmaceutical company CEOs.
• Search for Tesla as a company, then find recent news stories about them.
• Show me companies in the electric vehicle industry.
• Search for politicians mentioned in healthcare stories.
• What are tech companies saying about AI regulation?

Monitors:

• List my active event monitors.
• Show me the latest events detected by my product recall monitor.
• Create a draft monitor for executive departures in the semiconductor industry.
• Pause my competitor mentions monitor.
• Show me the latest newsletter from my AI regulation topic monitor.

Supported tools

The full list of available tools — including names, descriptions, and parameter schemas — is visible in the MCP playground. The tools available to you depend on the scopes granted to your API key.

Stats tools

The five stats tools are always available regardless of scope — the underlying /v1/stats/* endpoints perform no permission check beyond a valid key — and give aggregate metrics computed server-side, which is generally preferable to counting search results by hand.

Tool	Description
get_avg_sentiment	Average sentiment (positive/negative/neutral) bucketed over time for articles matching a filter set.
get_article_counts	Article publication volume bucketed over time — pair with get_avg_sentiment on identical filters for a trend view.
get_top_entities	The most frequently mentioned topics, people, companies, cities, journalists, or sources.
get_top_people	People whose coverage is spiking relative to a baseline period.
get_top_companies	Companies whose coverage is spiking relative to a baseline period.

Entitlement tools

Tool	Description
get_api_access	Always available. Reports this key's scopes, organization, usage quota, and the derived entitlement behavior (stripped fields, blocked filters, date-window clamps). Call once per session, or after a 403 or an unexpectedly null/empty field — does not count against request quota.

Monitor tools

The six read tools below are always available and expose the public /v1/api/monitors API for reading monitor configuration and output. create_monitor and update_monitor are not always-on — request them explicitly via ?tools=create_monitor,update_monitor or the monitoring profile, since most sessions never call them and the shared monitor schema is large.

Tool	Description
list_monitors	List and filter monitors by UUID, name, lifecycle status, or EVENT, MENTIONS, and TOPIC classification.
get_monitor	Retrieve a monitor's complete configuration, including its objective, query, output schema, schedule, watchlist, and contact points.
get_monitor_events	Retrieve structured events emitted by EVENT and MENTIONS monitors with extracted data, entities, summaries, and related articles.
get_monitor_newsletters	Retrieve scheduled human-readable briefings generated by monitors, typically for TOPIC monitors.
get_monitor_summaries	Retrieve rolling AI-generated monitor summary history.
set_monitor_status	Activate, pause, or archive a monitor. Archiving cannot be reversed through the public API.
create_monitor (opt-in)	Create a DRAFT or ACTIVE monitor with typed query, output schema, schedule, and delivery configuration. Defaults to DRAFT.
update_monitor (opt-in)	Partially update monitor configuration while preserving omitted fields.

Platform tools

Read-only by default; opt in via ?tools= (by name or the platform profile).

Tool	Description
get_source_by_id	Look up one news source by exact ID or domain for full detail (paywall status, alt names, location).
get_top_topics	Topics whose coverage is spiking relative to a baseline period.
get_story_stats	Story-level (clustered headline) publication volume or velocity over time.
watchlists	List, get by ID, or resolve by name your organization's watchlists of people and companies.
source_groups	List, get by ID, or resolve by name your organization's custom source-group bundles.
contact_points	List or get by UUID your organization's monitor notification delivery channels (email/webhook).
article_refresh	Check a background article-refresh job's status, or peek cached refresh data for up to 100 article IDs — read-only, cannot submit new jobs.
create_watchlist / update_watchlist	Create or partially update a watchlist. Only used when explicitly requested.
create_source_group / update_source_group	Create or partially update a custom source group. Only used when explicitly requested.

Research prompts

The server also registers six reusable MCP prompts that encode multi-step tool-chaining playbooks, so a host that surfaces prompts gets correct chaining without relying on the model to reconstruct it: entity_deep_dive, narrative_trace, coverage_trend, journalist_beat_profile, competitive_landscape, and spike_explainer.

Reference resources

Long-form guidance is available on-demand as MCP resources rather than always-on instructions: perigon://reference/fields (response field semantics), perigon://reference/chaining (cross-endpoint research playbooks), perigon://reference/entitlements (this session's scope-to-behavior mapping), and perigon://reference/charts (Signal Insights chart formatting rules).

Signal Insights tools

API keys with the Signal Insights scope unlock an additional set of tools for querying, exporting, and analyzing your AI signals data with a persistent Python sandbox.

Workspace pattern

Signal Insights tools use an explicit workspace handle (per SEP-2567):

1. Call create_insights_workspace once at the start of a conversation.
2. Pass the returned workspace ID to every subsequent analysis tool call.
3. Files written in execute_code or shell persist across calls within the same workspace. Exported data is accessible at /home/user/workspace/artifacts/ inside the sandbox.
4. If you resume a chat after a restart, the workspace UUID from the prior conversation is still valid — the sandbox kernel will be fresh but your exported S3 artifacts are preserved.

Signal Insights tool list

Tool	Type	Description
signal_insights_create_workspace	Setup	Create a workspace for the conversation. Must be called first for sandbox tools.
signal_insights_search_signals	Read	Search signals by name or objective; optional classificationTypes filter.
signal_insights_read_signal	Read	Get signal metadata (classification, schema or newsletter counts).
signal_insights_list_newsletters	Read	List newsletters for a TOPIC signal (title + excerpt).
signal_insights_read_newsletter	Read	Fetch full newsletter content as markdown.
signal_insights_export_events	Data	Export events to S3 (EVENT/MENTIONS only). Returns a preview and file path.
signal_insights_execute_code	Sandbox	Execute Python in a persistent IPython kernel. pandas, numpy, matplotlib and more pre-installed.
signal_insights_preview_chart	Sandbox	Render charts in the interactive chart viewer.
signal_insights_shell	Sandbox	Run bash commands in the sandbox.
signal_insights_list_files	Files	List files in the sandbox workspace.
signal_insights_read_file	Files	Read a file from the workspace.
signal_insights_write_file	Files	Write a file to the workspace.
signal_insights_grep	Files	Search file contents with a regex pattern.
signal_insights_str_replace	Files	Find and replace a string in a file.

Example config — Signal Insights only

{
  "mcpServers": {
    "perigon": {
      "url": "https://mcp.perigon.io/v1/mcp?tools=signal_insights_create_workspace,signal_insights_search_signals,signal_insights_read_signal,signal_insights_list_newsletters,signal_insights_read_newsletter,signal_insights_export_events,signal_insights_execute_code,signal_insights_preview_chart,signal_insights_shell,signal_insights_list_files,signal_insights_read_file,signal_insights_write_file,signal_insights_grep,signal_insights_str_replace",
      "type": "http",
      "headers": {
        "Authorization": "Bearer YOUR_PERIGON_API_KEY"
      }
    }
  }
}

Example config — Signal Insights combined with news tools

{
  "mcpServers": {
    "perigon": {
      "url": "https://mcp.perigon.io/v1/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer YOUR_PERIGON_API_KEY"
      }
    }
  }
}

With no ?tools= filter, all tools permitted by your API key's scopes are active.

Issues / Contributing

Issues

This MCP server is still in development as we determine what use cases our users want to solve with this server. But if you have any special requests or features you would like to see, don't hesitate to open a github issue on this Repo. We will gladly accept any feedback

Contributing

This tool is intentionally open source so if you want to see some particular feature you open an issue or open a PR and someone at Perigon will review it.

Local development

We are using bun for package mgmt.

Environment Variables

Add the following environment variables to .dev.vars

Variable	Description
ANTHROPIC_API_KEY	Anthropic API key (used for playground)
PERIGON_API_KEY	Perigon API key (used for playground)
POKEY_SIGNAL_INSIGHTS_BASE_URL	Internal Pokey service URL for Signal Insights MCP tools (e.g. http://localhost:3001). Required only when using Signal Insights tools.

If you wish to contribute to the MCP playground (tools inspector & chat), please make sure to modify your network hosts file (/etc/hosts on mac) to include the following

127.0.0.1 local-mcp.perigon.io

This will allow perigon.io cookies to be available for you while doing local development.

# install deps
bun i
# Runs the mcp server and the mcp playground
bun dev