Hosted setup guide

Trace Blox Docs

Get shared reasoning traces running for your whole team in a few clicks: a hosted MCP endpoint, with nothing to deploy or operate. Prefer to run it yourself? The free local-first and BYO-Cloudflare paths are further down.

Few-click team setupHosted MCP endpointLocal-first stays free

Evaluating first? How trace reuse works · the measured benchmark, with its scope · the threat model.

Cloud · hosted

Cloud-hosted setup

The fastest way to share reasoning traces across a team. Create a workspace, invite people, mint an API key, and point your agents at the hosted endpoint. No SQLite files, tokens, or Cloudflare Workers to manage yourself.

Create your cloud workspace

Sign in to the dashboard. Your first sign-in provisions a tenant and workspace for you. There is no infrastructure to stand up.

Open the dashboard →

Invite your team

Add members and assign seats from the Members panel. Everyone shares one hosted reasoning-trace store, so a trace one agent writes is instantly reusable by the rest of the team.

Members →

Create a cloud API key

In API keys, create a key and copy the one-time secret. Store it where your agent can read it; it is shown only once.

shell

export SCHOLIA_CLOUD_API_KEY="skl_live_…"   # the one-time secret from the dashboard

Point your agent at the hosted endpoint

Add the hosted Streamable-HTTP MCP endpoint to your client with the Bearer key. Nothing else about how you work changes.

Hosted MCP client configs ↓

Connect your agent: hosted MCP

Drop-in Streamable-HTTP configs for each client. They authenticate with the SCHOLIA_CLOUD_API_KEY you created above; no local process runs.

Claude

claude · hosted

{
  "mcpServers": {
    "scholialang-cloud": {
      "url": "https://mcp.traceblox.com/mcp",
      "headers": {
        "Authorization": "Bearer ${SCHOLIA_CLOUD_API_KEY}"
      }
    }
  }
}

Codex

codex · hosted

codex mcp add scholialang-cloud --url https://mcp.traceblox.com/mcp --header "Authorization: Bearer $SCHOLIA_CLOUD_API_KEY"

Cursor

cursor · hosted

{
  "mcpServers": {
    "scholialang-cloud": {
      "url": "https://mcp.traceblox.com/mcp",
      "headers": {
        "Authorization": "Bearer ${SCHOLIA_CLOUD_API_KEY}"
      }
    }
  }
}

Local-first · self setup

Local-first & self-hosted

Run everything yourself, for free. Local Scholialang writes to SQLite at ~/.scholialang/scholialang.sqlite3 by default; setting SCHOLIALANG_HOME lets each project keep its own local database and exports with no hosted account, bearer token, webhook, billing portal, or Cloudflare Worker.

Create a local storage root

shell

cd /path/to/project
mkdir -p .scholialang
export SCHOLIALANG_HOME="$PWD/.scholialang"

Keep raw traces out of Git

shell

printf '.scholialang/*.sqlite3\n.scholialang/*.sqlite3-*\n.scholialang/exports/\n' >> .gitignore

Install the local Codex plugin

shell

codex plugin marketplace add https://github.com/dougfirlabs/scholialang-mcp.git
codex plugin add scholialang@scholialang-mcp
codex plugin list

Smoke test the free local path

shell

codex
# Ask: Use Scholialang to start a local DAG, add a hypothesis, observation, evidence, and finding, then summarize the frontier.

Local data contract

The local store is a SQLite DAG: projects contain dags, dags contain atom nodes and relation edges, and summaries/counters can be generated from the same database. Cloud import preserves project, dag, atom, and edge identifiers so local work can remain the source of truth.

Connect your agent: local MCP

The same clients, over local stdio MCP. Swap to the hosted configs above whenever you move a project to the cloud; the tool surface is identical.

Claude

claude · local

{
  "mcpServers": {
    "scholialang": {
      "command": "python3",
      "args": ["/absolute/path/to/scholialang-mcp/plugins/claude-code/scholialang/scripts/scholialang_mcp_server.py"],
      "env": {
        "SCHOLIALANG_HOME": "/absolute/path/to/project/.scholialang"
      }
    }
  }
}

Codex

codex · local

codex plugin marketplace add https://github.com/dougfirlabs/scholialang-mcp.git
codex plugin add scholialang@scholialang-mcp
# Fallback: codex mcp add scholialang -- python3 /absolute/path/to/scholialang_mcp_server.py

Cursor

cursor · local

{
  "mcpServers": {
    "scholialang": {
      "command": "python3",
      "args": ["/absolute/path/to/scholialang-mcp/plugins/codex/scholialang/scripts/scholialang_mcp_server.py"],
      "env": {
        "SCHOLIALANG_HOME": "/absolute/path/to/project/.scholialang"
      }
    }
  }
}

Enterprise storage control

BYO Cloudflare

Self-host the storage plane: keep trace data in your own Cloudflare account while Traceblox Cloud runs the control plane.

Create a dedicated Cloudflare account or project with Workers, D1, R2, KV, Queues, and Durable Objects enabled.
Create D1 databases for control-plane routing and tenant trace storage, plus an R2 bucket for exports and attachments.
Create least-privilege API tokens for D1, R2, KV, Workers, and Queues. Store them as Worker secrets or CI environment secrets; do not commit them.
Bind SCHOLIA_CLOUD_DB, SCHOLIA_OBJECTS, SCHOLIA_KV, SCHOLIA_QUEUE, and Durable Object namespaces in Wrangler.
Run migrations from packages/db/migrations against the BYO D1 target, then verify tenant isolation with the tenancy gate before routing production traffic.
Use the dashboard import flow or CLI export/import to move reviewed local DAGs into the BYO tenant. Raw local exhaust can remain local.