Calyx MCP v1.0.5: Failure Recall Invariant & Persistence Error Propagation

Overview

Calyx MCP v1.0.5 hardens memory persistence and failure recall invariants, adds real stdio subprocess lifecycle testing, and expands the verification test suite to 51 passing tests.

---

What's Changed

🛡️ Failure Recall Invariant (Resolves PR #2)

• Pre-Truncation Failure Filtering: In ReflexDecisionEngine.evaluate_reflex(), failure records are now queried and evaluated before candidate truncation (failures_only=True). This ensures that clusters of subsequent high-similarity successes can never mask or overshadow an identified bug pattern from triggering the avoid reflex.
• Preserved Associative Ranking: Public associative queries (query_associative_memory) retain standard Hamming ranking across all outcome types.

💾 Persistence Error Propagation

• Strict Error Handling: MushroomBodyMemory.save_to_disk() now raises an OSError with diagnostic detail rather than silently swallowing I/O or filesystem errors. The MCP server catches this and returns a standard JSON-RPC error response (-32000) rather than falsely reporting status: "recorded".

🧪 Test Suite Expansion (51 / 51 Passing)

• Added tests/e2e/test_release_followups.py with 5 new regression and integration tests:
  • Failure recall resilience under multiple preceding and succeeding success records.
  • Synthetic write denial tests for weight saving and metadata atomic replacement.
  • End-to-end subprocess stdio JSON-RPC test verifying learning persistence across separate process instances.

---

Full Changelog: v1.0.4...v1.0.5

Calyx MCP v1.0.4: PyPI Banner & Documentation Fix

Overview

Calyx MCP v1.0.4 incorporates absolute asset URLs for package indexers (PyPI), strict API contract validation, Model Context Protocol (MCP 2024-11-05) tool annotations, an asymmetric contradiction resolution engine to eliminate false positives/negatives, synchronized bounded memory rings, and the 46-test verification suite.

---

What's Changed

1. Documentation & PyPI Asset Fix

• Resolved relative image path in README.md to use GitHub raw URL (https://raw.githubusercontent.com/ericmaddox/calyx-mcp/main/assets/calyx_banner.jpg), ensuring clean rendering on PyPI and package mirrors.
• Added comprehensive 46-test multi-tiered results table and complete MCP tool schema reference to README.md.

2. Robust Input Validation & Error Handling

• Strict outcome Enforcement: remember_code_outcome strictly requires outcome to be "success" or "failure". Unrecognized strings, typos, booleans, empty strings, and missing fields are rejected with JSON-RPC error -32000 before altering synaptic weights.
• Input Sanitization: Enforced non-empty string validation for code across check_code_reflex and remember_code_outcome, and non-empty query_code with integer-clamped top_k ([1, 50]) in query_associative_memory.

3. Standard MCP Tool Annotations

• Added official Model Context Protocol annotations across all 5 tools (readOnlyHint, destructiveHint, idempotentHint, openWorldHint).

4. Zero False Safety & Contradiction Resolution

• Failure Override Rule: Any matching episodic failure record (>= 0.65 similarity) unconditionally triggers status="avoid", preventing legacy positive history from masking newly discovered bugs.
• Contradiction Guard: Moderate failure resemblance (>= 0.50) alongside positive baseline valence safely returns status="neutral" with an explicit mixed-history warning.
• Recency-Prioritized Tie-Breaking: Similarity queries tie-break identical scores by newest record ID first.

5. Bounded Memory Ring Buffer

• Synchronized in-memory store and on-disk JSON registry with a strict FIFO capacity (max_records: 500) and monotonic sequence tracking (_record_seq).
• Added support for passive synaptic weight decay via apply_decay().

6. Multi-Tiered Verification Suite (46 Tests Passing)

• Unit tests for contradiction handling, polyglot resilience (Rust, TypeScript, Go, SQL, JSON), massive 150KB snippets, and memory lifecycles.
• E2E tests for API error contracts and 50-worker concurrent async execution.
• Token economics benchmark suite confirming sub-millisecond execution and token ROI.

---

Contributors

Special thanks to @rustybladerunner for identifying the remember_code_outcome validation edge case and proposing standard MCP tool annotations.

---

Installation & Upgrade

pip install --upgrade calyx-mcp==1.0.4

Calyx MCP v1.0.3: Hardened Validation, MCP Annotations & Contradiction Resolution

Overview

Calyx MCP v1.0.3 introduces strict API contract validation, Model Context Protocol (MCP 2024-11-05) tool annotations, an asymmetric contradiction resolution engine to eliminate false positives/negatives, synchronized bounded memory rings, and an expanded 46-test verification suite.

---

What's Changed

1. Robust Input Validation & Error Handling

• Strict outcome Enforcement: remember_code_outcome now strictly requires outcome to be "success" or "failure". Unrecognized strings, typos, booleans, empty strings, and missing fields are rejected with JSON-RPC error code -32000 before touching synaptic weights.
• Input Sanitization: Enforced non-empty string validation for code across check_code_reflex and remember_code_outcome, and non-empty query_code with integer-clamped top_k ([1, 50]) in query_associative_memory.

2. Standard MCP Tool Annotations

• Added official Model Context Protocol annotations across all 5 tools:
  • Read-only tools (check_code_reflex, query_associative_memory, inspect_memory_state): readOnlyHint=true, openWorldHint=false
  • State-mutating tools (remember_code_outcome, reset_memory): readOnlyHint=false, destructiveHint=true, idempotentHint=false, openWorldHint=false

3. Zero False Safety & Contradiction Resolution

• Failure Override Rule: Any matching episodic failure record (>= 0.65 similarity) unconditionally triggers status="avoid", guaranteeing that legacy rewards from older runs cannot mask newly discovered bugs.
• Contradiction Guard: Moderate failure resemblance (>= 0.50) alongside positive baseline valence safely returns status="neutral" with an explicit mixed-history warning.
• Recency-Prioritized Tie-Breaking: Similarity queries tie-break identical scores by newest record ID first.

4. Bounded Memory Ring Buffer

• Synchronized in-memory store and on-disk JSON registry with a strict FIFO capacity (max_records: 500) and monotonic sequence tracking (_record_seq).
• Added support for passive synaptic weight decay via apply_decay().

5. Multi-Tiered Verification Suite (46 Tests Passing)

• Unit tests for contradiction handling, polyglot resilience (Rust, TypeScript, Go, SQL, JSON), massive 150KB snippets, and memory lifecycles.
• E2E tests for API error contracts and 50-worker concurrent async execution.
• Token economics benchmark suite confirming sub-millisecond execution and token ROI.

---

Contributors

Special thanks to @rustybladerunner for identifying the remember_code_outcome validation edge case and proposing standard MCP tool annotations.

---

Installation & Upgrade

pip install --upgrade calyx-mcp==1.0.3

Calyx MCP v1.0.2: JSON-RPC 2.0 Notification Compliance & Antigravity Handshake Fix

What's Changed in v1.0.2

• MCP Notification Protocol Compliance: Properly ignores notifications/initialized and other JSON-RPC notifications without sending error responses on stdout, resolving handshake synchronization issues in Antigravity IDE and Cursor.
• Enhanced Parameter Tolerance: Supports both code and code_snippet parameter names across tools for wider MCP client compatibility.
• Test Validation: Full test suite passing across unit, integration, and STDIO protocol flows.

---

Installation

pip install calyx-mcp --upgrade

Full Changelog: v1.0.1...v1.0.2

Calyx MCP v1.0.1: Windows stdio streaming & Drosophila connectome visual

What's Changed in v1.0.1

• Windows I/O Streaming Hardening: Enhanced run_stdio() in server.py using thread executors to ensure robust, non-blocking JSON-RPC 2.0 streaming across all Python versions (including Python 3.13 on Windows).
• Executive Connectome Branding: Added the official 3D optical glass Drosophila melanogaster brain connectome banner to README.md.
• Packaging & Test Validation: Re-verified package distributions with twine check and full test suite (9/9 passed).

---

Installation

# Standard installation
pip install calyx-mcp --upgrade

# Zero-install instant run
uvx calyx-mcp

MCP Client Configuration

{
  "mcpServers": {
    "calyx": {
      "command": "calyx-mcp"
    }
  }
}

Full Changelog: v1.0.0...v1.0.1

Calyx MCP 1.0.0: Bio-Inspired Zero-Token Code Reflex Engine

Overview

Initial production release of Calyx MCP (v1.0.0), a bio-inspired associative memory and instant code reflex server for AI coding agents based on the Drosophila melanogaster (fruit fly) Mushroom Body circuit and Fly-LSH sparse projection algorithm.

Calyx delivers sub-millisecond pattern recognition and associative learning over the Model Context Protocol (MCP) without incurring LLM prompt token costs.

---

Key Highlights

• Zero-Token Fast Reflexes (<0.5 ms): Converts code AST tokens into 2,048-dimensional Kenyon Cell sparse representations (4.98% active sparsity) to detect known anti-patterns without LLM roundtrips.
• Dopaminergic Plasticity: Implements reinforcement and punishment synaptic weight updates ($\Delta W = \eta \cdot \text{Dopamine} \cdot (\mathbf{KC} \otimes \mathbf{MBON})$) to encode bugs or proven implementations.
• MCP 2024-11-05 Standard: 5 JSON-RPC 2.0 tools registered for code reflex checking, memory reinforcement, associative querying, memory inspection, and safe reset.
• Atomic Persistence: Synaptic states persist locally in compressed .npz format (~/.calyx/).
• Validated Performance: 9/9 automated tests passing across unit, integration, and STDIO protocol suites.

---

MCP Tools Included

1. check_code_reflex - Rapid pattern valence check (<0.5 ms, 0 LLM tokens).
2. remember_code_outcome - Dopamine-modulated synaptic plasticity update.
3. query_associative_memory - Fly-LSH nearest-neighbor similarity search.
4. inspect_memory_state - Mushroom Body operational metrics and weight distribution.
5. reset_memory - Memory reset safeguard.

---

Installation

pip install calyx-mcp

MCP Client Configuration

{
  "mcpServers": {
    "calyx": {
      "command": "python",
      "args": ["-m", "calyx_mcp.server", "--transport", "stdio"]
    }
  }
}