Model Context Protocol server for Agentic Task System — exposes the task app you already use to any MCP client, backed by hybrid + RRF retrieval. Storage-agnostic over the ATS adapter contract.
The ATS MCP server exposes 8 tools with basic schemas and descriptions, but quality is inconsistent across the suite. All tools have non-empty descriptions (10-200 chars range), and input schemas are present with parameter types and descriptions. However, parameter documentation is sparse, output schemas are completely undocumented, and descriptions lack the depth needed for reliable LLM selection. Most parameters lack constraints (enums, ranges, format specs), making it easy for LLMs to pass invalid values. Tool naming follows verb_noun convention but lacks discriminating detail, 'ats_find' vs 'ats_hybrid' vs 'ats_similar' all perform search operations and are not clearly differentiated in naming alone. No error handling guidance, no pagination support on list-like results, and no documentation of what each tool returns. The server reads from a pluggable adapter and supports hybrid search, but this architectural sophistication is not reflected in tool interface clarity.
Create a new task in the configured storage adapter
Search tasks across the configured storage adapter using hybrid retrieval (RRF parallel fan-out combining keyword and semantic search)
Retrieve a single task by ID from the configured storage adapter
Perform hybrid search combining keyword and semantic search using RRF fusion
Get related links and references for a task (wiki links, cross-references, related tasks)
Find tasks similar to a given task using semantic similarity
Update an existing task in the configured storage adapter
Output schemas completely undocumented. No tool documents what it returns, forcing LLMs to infer result structure and downstream parameter mapping. This violates pattern:tool and pattern:response-shaper.
Redundant search tools with unclear differentiation. 'ats_find', 'ats_hybrid', and 'ats_similar' all perform search-like operations but naming does not distinguish them, LLMs must read full descriptions to pick the right one. Violates pattern:tool (one clear job per tool) and review:name-clarity.
No pagination support. 'ats_find' and 'ats_hybrid' accept a 'limit' parameter but no offset/cursor/page_number. Large result sets will be truncated silently or blow context windows. Violates pattern:paginated-result.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Get the URL to a specific task in its native application
Parameter constraints missing. 'limit' parameters have no min/max bounds (baseline 1 - 100 or similar). 'query' parameters accept free-form strings with no length or character restrictions. LLMs can pass absurd values. Violates pattern:constrained-input and review:param-validation-rules.
No error handling guidance. No documentation of what errors a tool can raise, whether they are retryable, or what the LLM should do next. Violates pattern:recovery-guide and pattern:error-classification.
Tool descriptions lack LLM-optimization guidance. Descriptions are 55 - 70 characters, which is sparse. They state WHAT tools do but not WHEN to use them instead of similar tools, what they return, or any prerequisites. Violates pattern:tool-description and review:incomplete-docstrings.
'metadata' parameter on 'ats_create' is typed as 'object' with no schema, structure, or examples. LLMs will hallucinate field names. Should be a defined object with known fields or a reference to adapter-specific schema. Violates pattern:constrained-input.
'updates' parameter on 'ats_update' is typed as 'object' with no schema or constraints. LLMs cannot determine which fields are updatable, which are read-only, or what format values should take. Violates pattern:constrained-input.
No permission or scope documentation. Tools 'ats_create' and 'ats_update' are write operations but tool descriptions do not indicate required permissions. Violates pattern:scope-declaration and pattern:permission-gate.