MCP servers

The Model Context Protocol (MCP) lets AI agents call tools directly instead of parsing prose. HoneyCoin's docs are reachable through three MCPs — two are live today, one is in development.

At a glance

MCP	Status	When to use
ReadMe docs MCP	Live	Read or search HoneyCoin docs from inside an agent.
ReadMe authoring MCP	Internal	HoneyCoin team — edits to the docs themselves.
HoneyCoin product MCP	In development	Live route validation, coverage lookup, payload checks.

1. ReadMe documentation MCP

The hosted docs MCP exposes search and fetch over every public docs page. Use it when an agent needs to ground its output in the latest HoneyCoin guidance.

Tool	What it does
`search_docs`	Full-text search of HoneyCoin docs. Returns page slugs + snippets.
`fetch_page`	Retrieve the full Markdown of a page by slug.
`list_pages`	Sitemap — equivalent to `/llms.txt`.

When to use. Embed in Cursor, Claude Desktop, or any MCP-capable client. Configure it once and the agent can answer "how do I collect KES?" by pulling the reference page directly.

Configuration (Claude Desktop example):

{
  "mcpServers": {
    "honeycoin-docs": {
      "url": "https://docs.honeycoin.app/mcp"
    }
  }
}

2. ReadMe authoring MCP

Internal tool used by the HoneyCoin docs team. Exposes write operations on the docs themselves — creating pages, editing frontmatter, linting. Not intended for external integration builders.

Tool	What it does
`create_page`	Add a new page.
`update_page`	Edit existing page content.
`validate_links`	Audit cross-links and broken anchors.

This MCP is gated to the HoneyCoin engineering account and is documented here only to clarify what it is not for.

3. HoneyCoin product MCP (in development)

The product MCP will expose live HoneyCoin operations as tools, so agents can validate routes and payloads against the real platform — not just the static manifests.

Planned tool surface

Tool	What it does
`get_supported_routes`	List supported (country, currency, method, destination) combinations. Live equivalent of `coverage.json`.
`build_integration_plan`	Given a high-level goal ("collect KES, settle USDC on Base"), returns the step-by-step API calls + webhooks to subscribe to.
`get_endpoint`	Fetch the OpenAPI spec for a named operation by `operationId`.
`validate_payload`	Submit a request body for a named operation; returns whether it would be accepted, with field-level errors. Does not execute the call.
`get_webhook_schema`	Return the envelope and `data` shape for a named webhook event.
`get_test_data`	Sandbox fixtures (success, failure, OTP, redirect) for a given recipe.
`lookup_error`	Map an `errorCode` to its developer action, retry flag, and remediation.
`get_wallet_chain_support`	Wallet-specific coverage: chain → token → onramp/offramp/transfer support.
`build_wallet_integration_plan`	Step-by-step plan for wallet provisioning + webhook + reconciliation.
`validate_wallet_payload`	Like `validate_payload`, but for wallet endpoints. Catches invalid addresses, unsupported chain+token pairs, missing fields.

When to use the product MCP

You are an AI coding assistant building a HoneyCoin integration interactively. Use build_integration_plan to scaffold, then validate_payload before making real calls.
You are a developer using an agent to ask "is corridor X supported?" Use get_supported_routes.
You hit an error and want the agent to remediate. Use lookup_error.

Difference from manifests

	Manifests	Product MCP
Freshness	Snapshot at docs build time	Real-time
Access	Static JSON over HTTPS	Tool calls over MCP
Capability	Read only	Read + validate (no side effects in production)
Auth	None	Optional API key for account-specific support

When the product MCP ships, the manifests will remain the deterministic source of truth — the MCP will be a convenience layer on top.

Tool naming convention

All HoneyCoin MCP tools follow verbNoun naming to match the rest of the API (get_supported_routes, lookup_error, validate_payload). This mirrors the OpenAPI operationId convention.

Status and release

Milestone	Target
Spec finalized	Q2 2026
Public beta	Q3 2026
GA	Q4 2026

Agent instructions — Rules the MCP-driven agent must still follow.
llms.txt and skill.md — Static docs surfaces for agents.
Prompt examples — Prompts to test against an MCP-enabled agent.
Machine-readable manifests — Static data the MCP will eventually wrap.