Loading ToolBench

Reading the current corpus…

AHME-MCP MCP server · F · ToolBench

AHME-MCP

Asynchronous Hierarchical Memory Engine — a local sidecar daemon that compresses AI conversation context into a dense Master Memory Block via Ollama, SQLite, and asyncio.

<=2025-11-25STDIO3 tools

49/100

Definition Quality68

Protocol Readiness32

Supportability29

Spec posture

Inferred effective spec <=2025-11-25·current baseline 2026-07-28

Static source inference · medium confidence · detected: Logging

Deprecated protocol patterns detected

Summary

AHME exposes three well-named, verb-first tools (ingest_context, get_master_memory, clear_context) with clear descriptions (150 - 250 chars, within baseline 194 avg). All tools have input schemas with typed properties. However, critical gaps reduce the score: (1) output schemas are completely undocumented, callers cannot see what fields are returned; (2) parameter descriptions lack actionable constraints (e.g., no guidance on text length, reset behavior side effects, or namespace format); (3) no error handling guidance, tools return TextContent but do not document failure modes or recovery steps; (4) no per-parameter type validation hints (e.g., 'namespace must be 1-64 alphanumeric chars'); (5) destructive operations (clear_context, get_master_memory with reset=true) lack confirmation or dry-run patterns. Naming is strong (verb_noun convention, clear intent), but schema and error handling are weak.

Tools (3)

clear_contextdestructivesource verified64/100

Explicitly wipe all queued chunks and summaries from AHME's database. Use this if you want a completely clean slate without retrieving memory first.

get_master_memoryreversiblesource verified67/100

Retrieve the latest compressed Master Memory Block from AHME. Returns a dense, token-efficient summary of all ingested context, then resets the context window so the next session starts fresh with the summary as its seed (context-window replacement pattern). Inject the returned text into your system prompt to restore long-term memory.

ingest_contextwritesource verified72/100

Push raw conversation or document text into the AHME memory engine. AHME will asynchronously compress it into a hierarchical Master Memory Block.

Issues (5)

CRITICALtool ↗

Output schemas completely undocumented. Tools return TextContent but callers cannot see what fields, structure, or metadata are included in responses. LLMs cannot plan downstream tool calls or extract structured data.

ingest_contextget_master_memoryclear_context

HIGHconfirmation-request ↗

Destructive operations (clear_context, get_master_memory with reset=true) lack confirmation or dry-run patterns. Agents can irreversibly wipe memory without safeguards.

clear_contextget_master_memory

HIGHconstrained-input ↗

Parameter descriptions lack actionable constraints. 'text' has no length guidance; 'reset' does not explain side effects (DB cleared, re-seeded); 'namespace' has no format rules (alphanumeric? length limits?).

ingest_contextget_master_memoryclear_context

HIGHrecovery-guide ↗

Recommendations

Document output schemas for all three tools. Specify: (1) success response structure (e.g., {type: 'text', text: string, chunk_count?: number, tokens_used?: number}); (2) error response format (e.g., {type: 'text', text: string, error_code: string, retryable: boolean}).
Add parameter constraints to descriptions: 'text' → 'Raw text to ingest (1 - 1M chars, plain text or markdown)'; 'reset' → 'If true, clears DB after returning summary and re-seeds with summary for next session (irreversible)'; 'namespace' → 'Project namespace (1-64 alphanumeric + hyphens, auto-detected from AHME_NAMESPACE env or CWD name if omitted)'.
Implement confirmation pattern for destructive ops. Add optional 'confirm' boolean param to clear_context and get_master_memory (reset=true). Return {type: 'text', text: 'Confirm deletion of X chunks and Y summaries?', requires_confirmation: true} on first call; execute on second call with confirm=true.
Add error classification to tool descriptions and return values. E.g., 'On failure, returns {error_code: "namespace_not_found"|"db_locked"|"ollama_unavailable", retryable: true|false, message: string}'. Guide LLM: retryable errors → retry; user-fixable → ask user; fatal → abort.
Migrate from STDIO to HTTP+SSE or Streamable HTTP transport to enable remote deployment and hosted client compatibility. STDIO limits adoption to local-only setups.

Spec posture evidence

Inferred effective spec: <=2025-11-25.

Relies on Logging (deprecated) - log to stderr or use OpenTelemetry

Score history

Overall score trend

First recorded score · v2 rubric

49/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	49	<=2025-11-25	v2