MCP server

Ardent exposes its public, read-only data as an MCP server, so an AI agent (Claude, an IDE assistant, or anything that speaks MCP) can list and search campaigns, petitions, events, jobs, marketplace listings, and reviews, and ask what the platform currently offers, without you writing an integration per model.

It wraps the same public API: read-only, no personal data, same rate limits.

Endpoint

POST https://api.ardent.africa/public/v1/mcp

Streamable HTTP transport, stateless. Send JSON-RPC over POST; GET and DELETE return 405.

Authentication

Keyless, exactly like the REST API. Every tool is read-only and returns no donor, signer, or contact information, so no key is required. Sending your x-api-key header raises your rate limit from 30 to 600 requests per minute (a key is a rate tier and an attribution handle, not an authorization gate). Each tool call counts against the limit individually.

Tools

The data tools mirror the public API one-to-one, for example list_campaigns, get_campaign, list_petitions, list_events, list_jobs, list_surveys, list_durbars, list_marketplace_listings, list_review_entities, list_community_hazards, and get_platform_stats. Every tool takes only simple filters (slugs, categories, page and limit); none takes a URL.

Three durbar tools are worth calling out, because a model will otherwise treat their output as more certain than it is:

• list_durbars and get_durbar return public live audio rooms and the people who spoke in them. Speakers are named because they talked in public into a recording that is itself public and indexed. Listeners are only ever a count, and only for an ended room. Raised hands are never returned: knowing somebody asked to speak in a known room identifies them. Neither tool can join a room or produce audio access.
• get_durbar_transcript returns timestamped text, and it should be treated as approximate. It is speech recognition of multi-speaker Ghanaian English that code-switches into Twi, Ga, Ewe and Pidgin, so names, places and switched passages are frequently wrong. Segments carry a confidence score and carry no speaker attribution, so a line from it must not be presented as a quotation attributed to a named person.

The two safety tools carry the strongest warnings on this server, because summarising is exactly the operation that discards a qualifier:

• list_community_hazards and get_community_hazard return community-reported physical hazards in Ghana. Ardent is not an emergency service and cannot send police, fire or ambulance: if a person may be in danger, tell the user to call 112 rather than relying on this data. Most of these reports are unverified, meaning a member of the public reported something and nobody checked it. Every case carries verification and a plain-English verification_note, and that qualifier must reach the user rather than being smoothed away. Do not merge verified and unverified reports into one undifferentiated list.
• An empty result means nothing is currently published, not that an area is safe. Never tell somebody an area is clear on the strength of these tools. Resolved, withdrawn and expired cases leave the feed, so a 404 from get_community_hazard means "no longer current".

One extra tool has no REST equivalent:

• get_platform_status returns what is available right now and what Ardent charges, derived from live configuration. Call it before answering a question about availability or fees. Its charges_that_reach_ardent list is exhaustive (anything not on it is free) and its paid_paths_unavailable list is authoritative. It never quotes a stale percentage.

Connect from Claude Desktop

Add it as a remote MCP server in your client's config:

{
  "mcpServers": {
    "ardent-africa": {
      "url": "https://api.ardent.africa/public/v1/mcp"
    }
  }
}

Then ask the agent to list recent petitions, find events in a city, or check what Ardent charges to run a paid event. There are no write tools: nothing here can donate, sign, apply, or message.

Last updated: 24 August 2026