MCP Spec Checker

MCP Spec Checker

Semantic diff engine for MCP protocol specifications using three independent verification approaches with mandatory cross-validation.

Three Verification Approaches

GF(3) Conservation: (-1) + 0 + (+1) = 0 ✓

GF(3) Trit Assignments for Predicates

# From src/​mcp_spec_predicates.py

PREDICATE_TRITS = {
    # MINUS (-1): Constraint/​validation predicates
    "has_required_field": -1,
    "type_matches": -1,
    "schema_valid": -1,
    
    # ERGODIC (0): Coordination predicates
    "version_compatible": 0,
    "capability_negotiated": 0,
    "session_active": 0,
    
    # PLUS (+1): Generation/​action predicates
    "tool_invoked": +1,
    "response_emitted": +1,
    "resource_created": +1,
}

Denotation

This skill compares MCP protocol specs at the predicate level, detecting semantic differences between versions and generating minimal counterexamples for incompatibilities via cross-validated triadic verification.

SemanticDiff = Inv_0618 △ Inv_1125 (symmetric difference)
Counterexample: min{msg : Inv_0618(msg) ≠ Inv_1125(msg)}
Consensus: ∀ approach ∈ {Narya, Evaluators, Effects}: result_agree

Invariant Set

GF(3) Typed Effects

Narya Compatibility

Condensation Policy

Trigger: When 3 incompatible predicates are detected.

Action: Generate migration guide, emit counterexamples, mark as BREAKING.

Invariant Sets

Inv_0618 (June 2024 Spec)

Inv_0618 = {
    "initialize_required": True,
    "tools_list_before_invoke": True,
    "prompt_field_required": False,
    "result_or_error_exclusive": True,
    "capabilities_optional": True,
}

Inv_1125 (November 2025 Spec)

Inv_1125 = {
    "initialize_required": True,
    "tools_list_before_invoke": False,  # BREAKING CHANGE
    "prompt_field_required": True,       # BREAKING CHANGE
    "result_or_error_exclusive": True,
    "capabilities_optional": False,      # BREAKING CHANGE
}

Semantic Diff (Not Text Diff)

from mcp_spec_unified import semantic_diff

diff = semantic_diff(Inv_0618, Inv_1125)
# Output:
# {
#   "breaking": [
#     {"predicate": "tools_list_before_invoke", "0618": True, "1125": False},
#     {"predicate": "prompt_field_required", "0618": False, "1125": True},
#     {"predicate": "capabilities_optional", "0618": True, "1125": False},
#   ],
#   "compatible": [
#     {"predicate": "initialize_required", "value": True},
#     {"predicate": "result_or_error_exclusive", "value": True},
#   ],
#   "gf3_balance": 0  # Sum of predicate trits
# }

Counterexample Generation

When predicates disagree, generate minimal counterexamples:

from mcp_spec_unified import generate_counterexample

# Find minimal message that passes 0618 but fails 1125
counterex = generate_counterexample(
    spec_pass=Inv_0618,
    spec_fail=Inv_1125,
    predicate="prompt_field_required"
)
# Output:
# {
#   "message": {"jsonrpc": "2.0", "method": "tools/​call", "params": {"name": "example"}},
#   "0618_result": "PASS",
#   "1125_result": "FAIL: missing required field 'prompt'",
#   "fix": {"params": {"name": "example", "prompt": ""}}
# }

Cross-Validation (All 3 Approaches Must Agree)

from mcp_spec_unified import cross_validate

result = cross_validate(
    message={"jsonrpc": "2.0", "method": "tools/​call", ...},
    spec_version="1125"
)
# Output:
# {
#   "narya_types": {"result": "VALID", "trit": -1, "mode": "chk"},
#   "evaluators": {"result": "VALID", "trit": 0, "predicates_passed": 12},
#   "unison_effects": {"result": "VALID", "trit": +1, "effects_handled": ["IO", "Abort"]},
#   "consensus": True,
#   "gf3_sum": 0,
#   "confidence": 1.0
# }

Disagreement Handling

# If approaches disagree, report conflict
result = cross_validate(message, spec_version="1125")
if not result["consensus"]:
    print(f"CONFLICT: {result['conflicts']}")
    # Arbiter (ERGODIC) breaks ties
    final = result["evaluators"]["result"]

CLI Examples

# Compare two spec versions
just mcp-spec-diff 0618 1125

# Validate message against spec
just mcp-spec-check message.json --spec 1125

# Generate counterexamples for all breaking changes
just mcp-spec-counterex 0618 1125

# Cross-validate with all three approaches
just mcp-spec-validate message.json --cross-validate

# Run test trace
just mcp-spec-trace tests/​protocol_trace.jsonl

Test Traces

Valid 1125 Trace

{"seq": 1, "direction": "client->server", "message": {"jsonrpc": "2.0", "method": "initialize", "params": {"capabilities": {"tools": true}}}}
{"seq": 2, "direction": "server->client", "message": {"jsonrpc": "2.0", "result": {"serverInfo": {"name": "test"}}}}
{"seq": 3, "direction": "client->server", "message": {"jsonrpc": "2.0", "method": "tools/​call", "params": {"name": "example", "prompt": "test"}}}
{"seq": 4, "direction": "server->client", "message": {"jsonrpc": "2.0", "result": {"content": [{"type": "text", "text": "ok"}]}}}

Trace Validation Output

╔═══════════════════════════════════════════════════════════════════╗
║  MCP Spec Checker: Trace Validation                               ║
╚═══════════════════════════════════════════════════════════════════╝

Spec Version: 1125
Trace: tests/​protocol_trace.jsonl (4 messages)

─── Narya Types (chk/​syn) ───
  Message 1: ✓ chk(initialize) : Request
  Message 2: ✓ syn(result) : Response
  Message 3: ✓ chk(tools/​call) : Request
  Message 4: ✓ syn(result) : Response
  Trit: -1

─── Evaluators (GF(3)) ───
  Predicates: 12/​12 passed
  Breaking changes: 0
  Trit: 0

─── Unison Effects ───
  Effects handled: [IO, Abort, State]
  Unhandled: []
  Trit: +1

─── Cross-Validation ───
  Consensus: ✓ ALL AGREE
  GF(3) Sum: (-1) + 0 + (+1) = 0 ✓
  
RESULT: VALID

Source Files

Integration with Other Skills

Narya Type Modes

chk (checking mode):  Given type, check term has it
syn (synthesis mode): Given term, synthesize type
nosyn (no synthesis): Term cannot synthesize (must check)

Applied to MCP:

• chk(Request): Validate incoming message matches Request schema
• syn(response): Infer response type from message structure
• nosyn(partial): Partial messages require explicit type annotation

---

Skill Name: mcp-spec-checker
Type: Protocol Verification / Semantic Diff
Trit: 0 (ERGODIC - coordinates three approaches)
GF(3): Narya(-1) + Evaluators(0) + Unison(+1) = 0 ✓

Scientific Skill Interleaving

This skill connects to the K-Dense-AI/​claude-scientific-skills ecosystem:

Graph Theory

• networkx [○] via bicomodule
  • Universal graph hub

Bibliography References

• general: 734 citations in bib.duckdb

SDF Interleaving

This skill connects to Software Design for Flexibility (Hanson & Sussman, 2021):

Primary Chapter: 10. Adventure Game Example

Concepts: autonomous agent, game, synthesis

GF(3) Balanced Triad

mcp-spec-checker (+) + SDF.Ch10 (+) + [balancer] (+) = 0

Skill Trit: 1 (PLUS - generation)

Secondary Chapters

• Ch5: Evaluation
• Ch3: Variations on an Arithmetic Theme
• Ch4: Pattern Matching
• Ch7: Propagators

Connection Pattern

Adventure games synthesize techniques. This skill integrates multiple patterns.

Cat# Integration

This skill maps to Cat# = Comod(P) as a bicomodule in the equipment structure:

Trit: 0 (ERGODIC)
Home: Prof
Poly Op: ⊗
Kan Role: Adj
Color: #26D826

GF(3) Naturality

The skill participates in triads satisfying:

(-1) + (0) + (+1) ≡ 0 (mod 3)

This ensures compositional coherence in the Cat# equipment structure.