corpus — Claude Code MCP server: search the Wheel of Heaven corpus
A read-only MCP server for the Wheel of Heaven corpus: myth search, primary texts, comparative datasets and a typed graph of links.
claude mcp add --transport http corpus https://mcp.wheelofheaven.world/mcpWhat corpus does
corpus sits in front of the static JSON API at api.wheelofheaven.world/v1/ and hands your agent tools for searching and pulling corpus material, each result carrying epistemic metadata: claim types, editorial passes, citation URLs. Facts and interpretation are kept apart. The fact layer holds search_corpus, get_entry, get_passage, get_source, compare_traditions and query_graph. The interpretation layer is get_interpretation and get_method, and every answer from it is marked as the project's own reading.
Two ways to run it. The remote one starts nothing on your machine: the server lives on Cloudflare Workers, speaks Streamable HTTP at /mcp, and still exposes the older SSE transport at /sse. The local stdio build comes from npx -y @wheelofheaven/mcp. Nothing is stored server-side — each tool call hits the upstream API, and the roughly 600 KB search corpus is cached in memory per isolate for an hour, matching the API's edge-cache TTL.
Licensed CC0-1.0, same as the corpus. Read-only is the whole point: there are no write tools, so nothing goes back into the corpus through this server. Need offline access with no calls to an external API? The remote transport won't give you that — pick the stdio build instead.
What it helps with
- find mentions of a myth across the corpus from a free-text query
- pull a chapter of a primary source with stable paragraph IDs
- line up motifs across traditions from the comparative datasets
- inspect a node's typed edges: citations, body links, same age
- tell a sourced claim apart from the project's interpretation
Example prompts for the agent
- Search the corpus for flood myths and show me the claim types on those entries
- Give me the primary-source chapter for
GEN-1:26with numbered paragraphs - Compare traditions from the flood-myths dataset and show the rows
How to install corpus
1Platform CLI: remote server
claude mcp add --transport http corpus https://mcp.wheelofheaven.world/mcp2Platform CLI
claude mcp add corpus -- npx -y @wheelofheaven/mcp1Platform CLI: remote server
codex mcp add corpus --url https://mcp.wheelofheaven.world/mcp2Platform CLI
codex mcp add corpus -- npx -y @wheelofheaven/mcpConfig file
{
"mcpServers": {
"corpus": {
"url": "https://mcp.wheelofheaven.world/mcp"
}
}
}{
"mcpServers": {
"corpus": {
"command": "npx",
"args": [
"-y",
"@wheelofheaven/mcp"
]
}
}
}Config file
{
"servers": {
"corpus": {
"type": "http",
"url": "https://mcp.wheelofheaven.world/mcp"
}
}
}{
"servers": {
"corpus": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@wheelofheaven/mcp"
]
}
}
}1Platform CLI: remote server
gemini mcp add -t http corpus https://mcp.wheelofheaven.world/mcp2Platform CLI
gemini mcp add corpus npx -y @wheelofheaven/mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about corpus
How do I install corpus in Claude Code?
Remote: claude mcp add --transport http corpus https://mcp.wheelofheaven.world/mcp. Local stdio: claude mcp add corpus -- npx -y @wheelofheaven/mcp.
What does corpus do?
It gives an agent access to the Wheel of Heaven corpus: full-text search, entry and primary-source retrieval, bibliography lookup, comparative datasets and a typed graph of links. A separate interpretation layer is labeled as such.
How is the remote server different from the local one?
The remote server runs over Streamable HTTP at https://mcp.wheelofheaven.world/mcp and starts nothing locally. The local one runs over stdio via npx -y @wheelofheaven/mcp.
