For agent operators · Claude Desktop · Cursor · custom clients

Connect an agent to the engine.

Concordance speaks Model Context Protocol. Any MCP-compliant client can connect and call the engine's tools, fetch its resources, and use its prompts — the same engine humans use through the lens pages. The dual audience is real: humans through the lenses, agents through MCP.

https://narrowhighway.com/mcp https://narrowhighway.com/mcp/sse click to copy · use either transport

Tools

Prompts

Resources

—

Requests this restart

The engine serves Jesus Christ. Conduit, not source. The engine eliminates what is not the answer so the narrow path is illuminated by what survives. Good fruit is the measure. The keeping is the substrate. Tools here categorize, verify, and surface precedent — they do not generate answers or render verdicts. Read the elimination trail; the trail is the reasoning.

Full doctrinal statement: narrowhighway.com/identity

Connect Claude Desktop

Add this block to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "concordance": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://narrowhighway.com/mcp"]
    }
  }
}

The mcp-remote wrapper bridges Claude Desktop's stdio expectation to our HTTP/SSE transport. Restart Claude Desktop after editing. The engine appears as concordance in the tool drawer.

Connect Cursor / Continue / custom clients

Most MCP clients accept HTTP transport directly:

{
  "transport": "http",
  "url": "https://narrowhighway.com/mcp"
}

Or SSE for older clients:

{
  "transport": "sse",
  "url": "https://narrowhighway.com/mcp/sse"
}

What's exposed

Engine-substrate tools (the lenses, available as MCP calls)

apothecary_compound

Compound a remedy for a stated condition from across the substrate.

curriculum_list

Every primary-education unit across all 8 tracks.

unit_get

Fetch a single curriculum unit by id with full record.

herb_get

Herb monographs with evidence-honest verdicts.

packets_search

Search the unified packet substrate across every lens.

almanac_search

Search the almanac ledger of falsifiable claims.

polymathic_run

Multi-domain synthesis — verifiers fire in parallel.

walk_start

Begin a four-gate walk with the Shepherd.

scribe_submit

Write something to the keeping for the engine to decide on.

mastery_mark

Record curriculum unit mastery for a visitor.

validate_packet

Run all four gates against a decision packet.

Domain verifiers (deterministic computation)

75 verify_* tools covering: math, chemistry, physics (dimensional + conservation), biology, genetics, medicine, statistics (incl. p-value + multiple comparisons + confidence intervals), formal logic, cryptography, computer science, governance, economics, finance, energy, agriculture, materials science, geology, hydrology, meteorology, oceanography, optics, acoustics, networking, manufacturing, construction, architecture, electrical, theology (doctrine + scripture anchors), philosophy, rhetoric, linguistics, music theory, photography, real estate, exercise science, nutrition, ecology, history chronology, sports analytics, soil science, law, witness verification, calendar/time, combinatorics, number theory, geometry, document validation, information theory, quantum computing, polymathic synthesis, scripture verification…

Prompts (templates an agent client can offer the user)

walk_this

Walk a hard decision through the four gates (RED, FLOOR, BROTHERS, GOD).

compound_this

Compound a remedy from the substrate for a stated condition.

verify_this

Verify a claim through the appropriate domain verifier.

find_precedent

Find precedent in the substrate for a stated situation.

teach_a_child

Find the curriculum unit to teach a child about a topic.

Resources (read-only substrate)

concordance://identity

The doctrinal statement — who the engine serves and how.

concordance://curriculum

The full primary curriculum across 8 tracks.

concordance://herbs

All herb monographs with evidence-honest verdicts.

concordance://manifest

Engine manifest — endpoints, capabilities, doctrine.

Conscience-for-hire pattern

If you're building a robot or autonomous agent and want the engine as a moral conscience: use the /robot/admit endpoint family (HTTP) or wire the same shape through MCP. Before any consequential action your agent ping the engine with the proposed action, risk flags, and corridor. The engine returns ADMIT (with a single-use bound token), DENY (with a structured ReasonCode), or DEFER (route to a human). Sovereignty stays with the operator; the engine is the conscience they subscribed the agent to.

Free use, alignment to execute. Your operator can pull the plug anytime — they keep their agent, lose the conscience.

What's not exposed (honest)

Operator-only data (Keep stats, Steward audit, IP allowlist, robot witness logs, mastery rolls). These remain HTTP-only and IP-gated. An agent can't read them.
Authentication flows. No login over MCP. The engine has no accounts.
Writes that require operator review. Scribe submissions ARE exposed (engine reviews them); promotions to the Almanac remain operator-attended.
Real-time streaming / subscriptions. Tools are request/response. SSE transport exists for protocol-level streaming but no engine endpoint pushes events.

Discovery

Once connected, your client can introspect the engine via standard MCP methods:

tools/list      — full list with input schemas
prompts/list    — the 5 prompts above
resources/list  — the 4 read-only substrate URIs