Retrieve governed resources
Search and fetch approved, versioned CSI definitions, constructs, methods, and reference materials with stable identifiers and canonical citations.
Experimental research infrastructure
A machine-readable interface to CSI.
CSI MCP helps researchers and AI systems determine whether a proposed CSI analysis is theoretically coherent, sufficiently documented, and ready for empirical evaluation—before any collective-state inference is attempted.
Role in the research program
The machine interface—not the theory itself.
CSI MCP is a delivery and research instrument for interrogating the governed CSI record and constructing reproducible pre-inferential artifacts. It does not replace the theory, the canonical human-readable Reference, peer review, empirical validation, or accountable interpretation.
The current interface is deliberately algorithm-neutral. Its role is to help AI systems retrieve citable CSI materials, inspect whether proposed observations meet stated requirements, and preserve the reasoning chain that precedes any collective-state estimate.
Version 0.1 capability boundary
Governed access and pre-inference checks.
The prototype is intentionally narrow. “Read-only” means that it does not modify source records or take action on a collective; deterministic validation and ledger construction operate on submitted research inputs without producing a state estimate.
Validate observation sets
Check proposed evidence for boundary, time, level, provenance, privacy, and missingness requirements before inference is considered.
Build evidence ledgers
Apply explicit mapping rules while retaining provenance, competing evidence, unresolved questions, and a reproducible input record.
Practical research use
What researchers can do with it now.
CSI MCP turns parts of the CSI research protocol into a consistent, machine-checkable workflow. Its immediate purpose is to improve research readiness, traceability, and reproducibility—not to produce a collective-state result.
- Retrieve the authoritative construct definition and its stated evidence, composition, uncertainty, and qualification requirements.
- Submit a proposed observation set from a synthetic, pilot, or approved empirical study.
- Identify research-readiness problems involving collective boundaries, temporal alignment, level of analysis, provenance, privacy, and missing evidence.
- Produce a traceable evidence ledger that preserves mapping rules, supporting and competing evidence, unresolved questions, and reproducible input references for independent review or model comparison.
In an empirical study, those artifacts can support blinded review, preregistered mapping rules, comparisons among aggregate-only, relational, and CSI-based conditions, and an auditable record of how observations were prepared for analysis.
Evidence boundary: CSI MCP can support the production, governance, and auditability of empirical evidence, but it is not the source of that evidence and does not establish construct validity, ground truth, model performance, empirical advantage, or generalizability.
Explicit exclusions
What CSI MCP does not currently do.
- It does not infer, label, score, or assign probabilities to a collective state.
- It does not recommend interventions, issue alerts, modify permissions, or authorize action.
- It does not establish that CSI is empirically valid or that MCP is a standard implementation route.
- It is not suitable for personnel decisions, surveillance, monitoring, or other operational use.
These exclusions preserve the distinction between structured research support and a validated collective-state inference capability.
AI-system use
A governed way for AI systems to interrogate CSI.
For agentic systems, the interface could provide a consistent route from the CSI theory and Reference to machine-readable construct definitions, evidence requirements, composition assumptions, and qualification limits. That makes implementation claims easier to inspect and compare without assuming that tool access confers collective awareness.
CSI MCP does not establish that an AI-agent team possesses a collective state. The agent collective remains a proposed research domain requiring explicit constructs, appropriate composition logic, uncertainty, competing explanations, and collective-level validation.
Development path
From private scaffold to evaluated reference interface.
Current
Private development scaffold
Local retrieval, observation-set validation, and evidence-ledger construction are implemented against controlled resources and fixtures.
Next
Independent evaluation
Complete blinded review, baseline comparisons, burden analysis, disagreement handling, and release-readiness decisions.
Later research direction
CSI-conformant specifications
If governance and evaluation requirements are met, later versions may support inspectable CSI-conformant inference specifications. This is not a current capability.
Access status: The source repository, technical documentation, and development endpoint remain private during v0.1 review. Public deployment, packaging, and licensing decisions are pending.
Research collaboration
Interrogate the interface and its boundaries.
Independent feedback is especially useful on tool semantics, evidence provenance, evaluation design, competing baselines, governance constraints, and what would falsify the claimed value of a machine-readable CSI interface.
Contact Edward ClarkView the Research Program