Loading ToolBench
Model Context Protocol integration for Gravity Well — continuity infrastructure for AI agents. Provides tools for registration, identity bootstrapping, content capture, encryption, deposit to Zenodo, and state reconstitution with compression-survival scoring.
Gravity Well MCP server has well-structured tool definitions with consistent naming (gw_* prefix) and detailed descriptions. However, critical gaps undermine production readiness: (1) Input schemas are present but lack explicit type definitions for nested objects (e.g., bootstrap_manifest, context_anchors are typed as 'object' without property schemas); (2) Parameter descriptions are verbose but often domain-specific jargon ('glyphic checksum', 'Caesura', 'SIM injection') that LLMs cannot reliably interpret without external context; (3) Output schemas are completely undocumented, no tool specifies what fields it returns, forcing LLMs to guess at response structure; (4) Error handling is absent, no tool describes failure modes, recovery paths, or actionable error messages; (5) Security concerns: api_key is exposed as a required parameter in every tool, violating secret-injection pattern. The domain (continuity chains, glyphic encoding, stratified compression) is novel and internally consistent, but lacks the clarity and structure production agents require.
Generate an identity manifest (bootstrap) from name, description, and constraints. The manifest defines who the agent is and what it must never do. No auth required.
Capture content to a chain. This stages the content for later deposit. Use visibility 'public' for readable content, 'private' for sensitive content (stored as-is — use Python encryption client-side before calling if needed). Optional glyphic_checksum for structural topology encoding.
List all provenance chains for your API key. Returns chain metadata including labels, DOIs, versions, and auto-deposit settings.
Continuity console for a chain. Returns full chain state: staged objects, deposit history, bootstrap manifest, auto-deposit settings, and latest glyph trajectory.
Create a continuity chain. The chain stores all captures and deposits. Use anchor_policy 'local' for private chains (no DOI), or 'zenodo' for public DOI-anchored chains. Optionally include a bootstrap_manifest to store identity on the chain permanently.
Output schemas completely undocumented. No tool specifies what fields it returns (e.g., gw_register returns an API key, but the response structure is not documented). LLMs cannot plan downstream tool calls without knowing response shape.
API key exposed as required parameter in all 15 tools. Violates secret-injection pattern, credentials in tool parameters are logged in traces and prompt history. Must use server-side secret injection via environment or vault.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Wrap and deposit staged content to Zenodo (or local storage if anchor_policy='local'). Applies Evidence Membrane, Caesura, SIM injection, and Integrity Lock. Returns DOI if Zenodo deposit.
Check identity drift — compare current bootstrap manifest hash against stored hash on chain. Returns drift status and constraint violations if any.
Score compression survival (γ) of content. High γ = content survives LLM summarization with referential integrity intact. Analyzes citation density, structural integrity, argument coherence, provenance markers, information density, redundancy, argument chain depth, citation integration, and vocabulary specificity.
Generate Ledger deposit (stratified continuity compression). Creates a compressed summary of an Archive chain and links it bidirectionally. Ledger is a Tier 2 compression layer for long-running chains.
Recover four-layer state package from a chain: bootstrap (identity), tether (operational handoff), narrative (structural summary), glyphic trajectory (glyph sequence), and provenance (DOI chain).
Create a Gravity Well API key. Returns a key that must be saved — it cannot be retrieved again. Use this first before any other GW operation.
Retrieve context anchors from a chain. Returns stored glyphic context sequences for glyph trajectory reconstruction and structural continuity.
Retrieve and decrypt encryption key from Supabase. Returns the decrypted key material (as base64). Requires KEK to be available in environment.
Store glyphic context anchors (Tier 2 — stratified continuity compression). Stores emoji-encoded structural context for later retrieval and glyph trajectory reconstruction.
Store encryption key in Supabase (encrypted with KEK — Key Encryption Key). The key is encrypted before transmission and storage. Requires Supabase configuration.
Nested object parameters lack property schemas. gw_bootstrap 'constraints' is array of strings (good), but gw_create_chain 'bootstrap_manifest' and gw_store_context 'context_anchors' are typed as 'object'/'array' without defining internal structure. LLMs cannot validate or construct these payloads.
Domain-specific jargon in descriptions without explanation. Terms like 'glyphic checksum', 'Caesura', 'SIM injection', 'Evidence Membrane', 'stratified continuity compression' are not standard LLM vocabulary. Descriptions must be self-contained and avoid unexplained domain terminology.
No error handling guidance. Tools do not describe failure modes, recovery paths, or actionable error messages. E.g., gw_register says 'Returns a key that must be saved, it cannot be retrieved again' but does not explain what happens if registration fails or how to recover.