Assessment MCP Server

An assessment platform for ADK-Rust Enterprise education agents. 23 MCP tools covering an item bank (six item types with keyed answers), standards/curriculum mapping, assessment authoring + blueprint generation, rubric scoring, attempts with auto-grading, and psychometric analytics (item difficulty/discrimination and standard mastery) — with a full audit trail.

A platform, not a point solution

This is modeled as a general assessment engine (à la a test/item-banking system behind an LMS), so the education agents are simply clients:

| Agent | Uses | |-------|------| | Quiz Generator | generate_assessment (difficulty blueprint over the item bank), create_item | | Curriculum Mapping Agent | add_standard, list_standards, items_for_standard, assessment_blueprint | | Instructor Copilot | assessment_blueprint, item_analysis, list_attempts, set_assessment_status | | Rubric Grading Assistant | create_rubric, score_rubric_item | | Adaptive Tutor Agent | standard_mastery, learner_attempts, items_for_standard |

Architecture

Capabilities

• Item bank — six item types: multiple_choice, multiple_select, true_false, short_answer, numeric (with tolerance), and essay (rubric-scored). Items carry difficulty, Bloom's cognitive level, point value, and standard alignment. Authoring is validated per type.
• Standards / curriculum mapping — learning standards (CCSS/NGSS/custom) with items_for_standard coverage and per-assessment standard breakdowns.
• Authoring & generation — build assessments from explicit items, or generate from a blueprint (difficulty → count, optionally constrained to standards) drawing from the bank.
• Attempts & grading — learner attempts auto-grade objective items on submit; rubric/essay items are scored per-criterion (clamped to each criterion's max) and the attempt finalizes with score, percent, and pass/fail.
• Psychometrics — item_analysis computes per-item p-value (difficulty) and discrimination (top-third vs bottom-third correct-rate) with quality flags (too_hard / too_easy / low_discrimination). standard_mastery rolls a learner's earned points per standard into proficiency levels.

Governance posture

• Two writes are gated (requires_approval): publishing an assessment (set_assessment_status → published — makes it live for learners) and submitting an attempt (submit_attempt, classed external_write — it records a graded result of record).
• Publishing is guarded — an assessment can't be published with no items, and only published assessments accept attempts.
• All analytics and bank queries are read_only.
• Everything is audited (audit_log). Sample data is fictitious.

Tools (23)

Standards & Curriculum (3)

add_standard · list_standards · items_for_standard

Rubrics (2)

create_rubric · get_rubric

Item Bank (3)

create_item · get_item · list_items

Assessments (6)

create_assessment · generate_assessment · get_assessment · list_assessments · set_assessment_status (gated) · assessment_blueprint

Attempts & Grading (6)

start_attempt · submit_attempt (gated, external) · score_rubric_item · get_attempt · list_attempts · learner_attempts

Analytics (3)

item_analysis · standard_mastery · audit_log

Example

// Quiz Generator: build from a difficulty blueprint
{"name": "generate_assessment", "arguments": {"title": "Algebra Quiz", "subject": "Mathematics",
  "blueprint": [{"difficulty": "easy", "count": 3}, {"difficulty": "medium", "count": 2}]}}
{"name": "set_assessment_status", "arguments": {"assessment_id": "ASM-1012", "status": "published"}}

// Attempt + auto-grade
{"name": "start_attempt", "arguments": {"assessment_id": "ASM-1012", "learner_id": "L1"}}
{"name": "submit_attempt", "arguments": {"attempt_id": "ATT-1020",
  "responses": [{"item_id": "ITM-1001", "value": 4}]}}

// Instructor Copilot + Adaptive Tutor
{"name": "item_analysis", "arguments": {"assessment_id": "ASM-1012"}}
{"name": "standard_mastery", "arguments": {"learner_id": "L1"}}

Install & run

cargo install mcp-assessment
mcp-assessment            # serves MCP over stdio

Or build from source:

git clone https://github.com/zavora-ai/mcp-assessment
cd mcp-assessment && cargo build --release
./target/release/mcp-assessment

Registry manifest

server_id = "mcp_assessment"
display_name = "Assessment Platform"
version = "1.0.0"
domain = "education"
risk_level = "medium"
writes_allowed = "gated"

The full mcp-server.toml declares all 23 tools with risk classes and approval gates for registry onboarding.

License

Apache-2.0