A 1C:Enterprise native extension component that implements an MCP server over HTTP, supporting semantic/keyword/hybrid search over indexed segments via a Rust core (rcore), with native tools for search, grep, segment retrieval, and collection management. Integrates with 1C backend for tool/resource/prompt registry synchronization.
The server implements 4 well-named search and data retrieval tools with READ_ONLY semantics. Tool names follow verb_noun conventions (search, get_segment, grep, list_collections) and are appropriately action-oriented. However, definition quality is uneven: the 'search' tool has a comprehensive, well-typed input schema with enum constraints and clear descriptions, but 'get_segment', 'grep', and 'list_collections' lack visible input schema definitions in the provided source code. Descriptions are present but brief (10-60 chars), below the production baseline of 194 chars. Output schemas are not documented. Error handling is present in the C++ implementation (e.g., via json::parse error checking), but recovery guidance is minimal. The server demonstrates good naming discipline and some parameter validation (enums for 'mode' in search), but falls short of production-grade definition rigor due to schema gaps and sparse descriptions.
Retrieve a specific segment by its ID.
Regex-based search using RE2-style finite automaton matching (no catastrophic backtracking).
List available search collections.
Semantic/keyword/hybrid search over indexed segments. Returns ranked hits with doc_id, name, score and segment line ranges.
Input schemas missing or not visible for 3 of 4 tools (get_segment, grep, list_collections). Per hard scoring rules, tools without visible input schemas receive schema score = 0.
Tool descriptions are extremely brief (28-42 chars for 3 tools). Production baseline is 194 chars (p10=34, p90=392). Descriptions lack context on WHEN to use the tool, WHAT it returns, and HOW it differs from similar tools. E.g., 'get_segment: Retrieve a specific segment by its ID' does not explain segment semantics, what fields are returned, or whether segment IDs come from search results.
Output schemas are not documented for any tool. LLMs need to know what fields to expect (e.g., search returns 'doc_id, name, score, segment_line_ranges' per description, but the full structure is not documented). Without response schema, agents cannot reliably extract returned values for downstream tool chaining.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
grep tool description mentions 'RE2-style finite automaton' but does not document the expected parameter (regex pattern string, presumably), constraints (length limit? special chars allowed?), or whether the pattern accepts flags. LLM must guess the API.
Error handling is defensive in C++ (e.g., parseJsonSafe, parseJsonOrNull) but does not propagate recovery guidance to the LLM. When a segment ID is not found or a regex fails, the response should offer next steps ('Try search() to find valid segment IDs') rather than a bare error. Per pattern:recovery-guide, errors must tell the agent what to do next.