A research graph your AI can inspect, challenge and act on

This is a Model Context Protocol server over the Human Repair Graph: what humanity can and cannot yet do to repair a human body, as goals → capabilities graded on demonstrated evidence → open questions → claims → sources. Two public maps (universal repair, rejuvenation) are projections of one graph. 185 capability, 31 cell, 24 claim, 6 experiment, 10 goal, 20 question, 16 route, 34 source. Content hash 6c6db5f42d33.

Add it to any MCP client (Claude Desktop, Claude Code, or your own agent):

claude mcp add --transport http human-repair-map https://humanrepairmap.com/mcp

Or in a client config:

{
  "mcpServers": {
    "human-repair-map": { "type": "http", "url": "https://humanrepairmap.com/mcp" }
  }
}

Tools

• graph_manifest — Version, snapshot date, content hash, node counts by type and review state, bulk-export URLs and the caveat.
• how_to_read — The evidence ladder (L0–L5), the grounding ladder (G0–G4), what 'measured' means, the capability classes, repair primitives, grade bases and review states.
• get_node — Full record for any node by id (goal, capability, question, claim, experiment, source, cell, route) with its derived inverse relations..
• list_nodes — Filter nodes by type, projection, capability class, rung, grade basis, blocked kind, state, review state, or free text.
• search — Free-text search across every node type (and the v0.1 routes and capabilities)..
• get_subgraph — Breadth-first neighbourhood of a node over chosen relations (requires, enables, blockedBy, blocks, supportedBy, contradictedBy, testedBy, parent, children, target) to a given depth.
• trace_dependency — For a goal: every capability it transitively requires, the binding constraints (AND-required capabilities at the lowest rung), the OR groups where one alternative suffices, the ungraded ones, and the open questions along the path.
• find_blockers — The open questions that gate a goal or capability, directly and through everything it requires..
• rank_research_questions — Open questions ranked by structural leverage: how many goals and capabilities sit downstream of what each one blocks, tie-broken by how low the blocked capability sits.
• get_primary_evidence — The claims supporting or contradicting a capability (or the claim itself), each with its sources: DOI/PMID/URL, study design, locator, whether a machine resolved the source and whether a human has opened it..
• find_contradictions — Claims that contradict a capability or another claim — including null results.
• what_would_move_this — For a capability, claim, route or question: the specific missing demonstration that would raise its rung or resolve it, with the drift between demonstrated and claimed..
• register_prediction — Lock a forecast about a question, claim or capability BEFORE the outcome is known.
• list_predictions — Predictions registered against a node (or all), with their state..
• propose_change — File a correction, rung challenge, new evidence, new node, new dependency or question against any record.
• evidence_state — The headline state of getting a drug into the human brain: routes by evidence rung, what is proven in humans, the central contrast..
• repair_grid — Every human cell type plus the six non-cellular things, graded on see/model/reach/edit/verify.
• get_cell — Full record for one cell type or non-cellular node..
• list_routes — The 16 graded routes into the human brain, filterable..
• get_route — Full route record..
• list_capabilities — The five enabling capabilities with maturity scores..
• get_capability — Full v0.1 capability record..

Every tool returns structuredContent (JSON to reason over) alongside a text rendering, and every payload carries the review state of what it returned.

Without MCP

• REST — /api (index) · /api/openapi.json. Every node is a URL: /api/nodes/hrm:goal/scarless-skin-repair, /api/goals/scarless-skin-repair/critical-path, /api/questions/ranked.
• Bulk — /graph/graph.json · graph.jsonl · graph.jsonld · manifest.json · JSON Schemas. Immutable per content hash.
• Act — POST /api/proposals files a correction; POST /api/predictions locks a forecast against the snapshot you saw. Both land in the public hash-chained log (/api/verify).

Spec 2025-11-25 · Streamable HTTP · humanrepairmap.com · source