exa-mcp-server Releases

This release focuses on security improvements, introduces enterprise PLC reader/writer blocks, and adds support for a vLLM proxy backend. It also includes fixes for SAM2 caching and various workflow enhancements.

Jun 26, 2026

This release adds a Switch Case flow-control block and improves runtime introspection for Jetson devices. It also fixes keypoint visualization issues and enables loading locally stored models in workflows.

Jun 19, 2026

A maintenance release focused on SAM 3 improvements including streaming video concept tracker, point prompting capabilities, and various workflow enhancements. Also includes WebRTC SDK improvements and serverless support for SAM 3.

Jun 12, 2026

A release focused on adding new database connectors and improving existing ones for Pathway. Introduces Elasticsearch, ClickHouse, and MySQL connectors with various enhancements to Iceberg, MongoDB, and PostgreSQL connectors.

Jun 12, 2026

This release introduces significant enhancements to the inference engine, including support for RF-DETR Keypoints and fine-tuned YOLO26 Semantic Segmentation models. It adds new workflow blocks for dynamic time injection and local Vision Events, alongside improvements to existing blocks like GLM-OCR and Qwen3.5-VL to support dynamic parameter selection. The release also includes comprehensive security hardening documentation for self-hosted deployments and various stability fixes.

Jun 5, 2026

This maintenance release focuses on resolving bugs related to semantic segmentation execution and model aliases while updating the default Gemini model version.

Jun 2, 2026

Adds SQLite write connector, improves Delta Lake data type support, and fixes Delta Lake partition decoding.

May 25, 2026

Security patches addressing RUSTSEC advisories in `rustls-webpki` and `wasmtime`, plus maintenance updates and fixes carried forward from v0.6.8.

May 12, 2026

Focuses on workspace isolation, dashboard authentication, agent tools, and local provider overrides.

May 12, 2026

Improvements to the web dashboard (reconnect, uninstall), agent lifecycle (stop, activate), and runtime environment.

May 12, 2026

Updates to LLM drivers, Discord image support, and new tools.

May 12, 2026

Agent lifecycle management (wakeup, clone) and dashboard improvements.

May 12, 2026

This release introduces several new I/O connectors for RabbitMQ, SQL Server, Milvus, and LEANN, along with an audio transcription parser and improvements to multi-machine deployments and operator persistence.

Apr 23, 2026

This release introduces new database connectors for MongoDB and PostgreSQL, along with enhancements for array serialization and Airbyte dependency management. It includes breaking changes regarding MongoDB array shape preservation and dependency inclusion for Python filesystem connectors.

Mar 24, 2026

This release introduces OAuthBearer authentication for Kafka connectors, adds snapshot mode support to MongoDB writes, and enables automatic worker scaling based on pipeline load. It also improves TLS configuration for Postgres writes and enhances connection retry logic.

Feb 16, 2026

This release introduces new model variants (gemma3 and qwen3), improves language understanding (specifically Hindi), enhances robustness against noise, and adds a new evaluation suite called VoiceBench.

Aug 18, 2025

v0.5 brings significant performance improvements with 60% better transcription accuracy, 24% better translation, and expanded language support from 15 to 42 languages. The weights have been pushed to Hugging Face and are the new default for Realtime APIs.

Feb 11, 2025

v0.4.1 upgrades the Whisper encoder to Whisper-large-v3-turbo, resulting in improved translation quality across multiple language pairs. It adds six new languages to the existing 15 and increases the size of the English training data.

Nov 12, 2024

v0.4 introduces a major upgrade to the Whisper encoder and expands training from English-only to a multi-lingual dataset, supporting 9 languages. It uses a frozen Llama 3.1 8B core and trains on roughly doubled data size.

Aug 27, 2024

v0.3 demonstrates substantially improved speech understanding by using a frozen Llama 3.1 8B core and a 2.5k-hour speech adapter. It introduces synthetic data augmentation and Knowledge Distillation for better alignment, alongside access to managed APIs.

Aug 23, 2024