CasimirBot / Helix

Agent Access

Connection information for the provider-neutral CasimirBot external-agent harness.

An explicit connection is required

A model that retrieves this public page can read and cite it, but retrieval alone cannot invoke CasimirBot tools. An MCP-capable host must configure an endpoint; the account MCP endpoint requires OAuth authorization.

Credentials belong in the client's OAuth connection. Do not put access tokens or secrets in prompts or URLs.

Reason with evidence while disconnected

An AI client can use this public brief to structure its own answer before an MCP connection exists.

Identify the user's actual question, requested source, and intended action. Quoted, negated, historical, or hypothetical tool words are not commands.
Keep source facts, tool observations, and your own inferences distinct. Name the source and uncertainty for each material claim.
When an authorized MCP connection exists, inspect the admitted read tools and re-evaluate their returned evidence. A route proposes a procedure; a receipt records an observation. Neither alone is the final answer.
Check missing evidence and contradictions before concluding. If the available evidence cannot support the requested conclusion, say what remains unverified.

This public brief is guidance, not a live Helix reasoning trace. Do not claim a CasimirBot run state, tool result, proof verdict, or terminal decision without its authenticated, typed evidence.

Use public CasimirBot context in your own AI chat

If an AI host retrieves this page, it can use the public context without an account or MCP connection. The host chooses whether to follow a pasted link and remains responsible for its answer.

Civilization Bounds: For society and collective-action questions, examine affected people, social and material dependencies, collaboration, evidence gaps, and ways a proposal could fail.
Theory Badge Graph: For scientific claims, trace assumptions, mechanisms, and evidence boundaries before treating a model or simulation as support.
Moral Graph: For questions about welfare or obligation, identify affected beings and make value judgments and uncertainty explicit.

A broad question about society may benefit from all three lenses even if the user names none of them. Apply only the relevant lenses; this page does not authorize a tool call or establish real-world facts.

Read the public reasoning context and source notes before applying a lens. Cite independent evidence for claims about the world.

Explore the public badge catalog for concept details, uncertainty, source references, and claim boundaries.

Connect from an AI host

In the AI host's MCP or connector settings, add https://casimirbot.com/mcp as a remote Streamable HTTP server. A host that only reads web pages can use this guide but cannot call the tools.
Sign in to the CasimirBot account through that host's OAuth flow and review the requested scopes. Keep credentials in the host connection, never in the prompt.
Use the host's authenticated tool list to see what this account and connection can call. Free personal read access and hosted Helix Ask runs are separate grants. Hosted run start and continuation currently require developer access.

ChatGPT, Codex, OpenAI API, and Gemini configuration examples are in the Agent API and MCP contract.

Free public context tools

An MCP-capable host may separately configure https://casimirbot.com/mcp/public-context without an account. It offers only casimir_civilization_bounds_reflect, casimir_theory_badges_reflect, and casimir_moral_graph_reflect. Each returns bounded read-only context for one question. The host forms the answer. Pasting a link does not automatically attach MCP tools.

Use a free personal account

After OAuth connection, start with the read-only helix_public_ui_catalog tool or another tool admitted by the authenticated catalog. The AI host reads the tool receipt as supporting information and writes its own response. CasimirBot does not run a hosted model for this read-scoped path.

Two read-scoped starter tools require helix.agent_runs.read:

helix_workstation_capability_list
helix_public_ui_catalog

Tool access depends on the account and granted scopes. A pasted link, a public page read, or an installed desktop harness session does not authorize the AI host. Hosted run start and continuation are separate developer capabilities until a customer model-funding gate is available.

Service endpoints

Surface	Canonical URL	Purpose
MCP	`https://casimirbot.com/mcp`	OAuth-protected, stateless Streamable HTTP MCP requests.
Durable-run REST API	`https://casimirbot.com/api/v1/agent-runs`	OAuth-protected durable run lifecycle.
Shared Live Room REST API	`https://casimirbot.com/api/v1/rooms`	OAuth-protected room, run-binding, and deferred source control.
OAuth metadata	`https://casimirbot.com/.well-known/oauth-protected-resource/mcp`	Protected-resource discovery for authorization clients.
API contract	`https://casimirbot.com/docs/architecture/helix-agent-api-v1.md`	Run lifecycle, schemas, scopes, and safety boundaries.
Workstation	`https://casimirbot.com/open?panels=agent-access&focus=agent-access&entry=workstation`	Human-facing connection and configuration surface.
CasimirBot metadata	`https://casimirbot.com/agent-access.json`	Machine-readable links from this page; not a standards-defined manifest.

Core durable-run MCP tools

These are the stable run-lifecycle tools; hosted run start and continuation currently require developer access. The authenticated MCP tools/list response is the source of truth for the complete catalog, including Shared Live Room tools. Each tool declares its complete required OAuth scope set, which is enforced when called.

helix_run_start
helix_run_evidence_reenter
helix_run_continue
helix_run_inspect
helix_run_fetch_evidence
helix_run_list_events
helix_run_cancel

Durable run model

Clients start a bounded run, retain its opaque run_id, and continue or inspect that same run. Results include structured status, evidence references, unresolved requirements, and recommended next actions.