v0.0.21

Minor patch update to get mcp-agent compatible with latest MCP package v1.8.0

Changes

• Upgrade MCP package dependency to v1.8.0 with corresponding hotfixes (8de63b4)
• Add an OllamaAugmentedLLM and ensure it works well with complex workflows that require structured type generation (
• Fixing absolute path of server in basic/decorator/yaml @Deathn0t (#181)
• Get rid of custom websocket and stdio clients in favor of default ones shipped in MCP package (9d92ef0)

v0.0.18

Thank you to all the wonderful community contributions! We are gearing up for a significant update to mcp-agent shortly. You can follow the progress in feature/temporal_prime and feature/app_server branches. v0.0.18 has a few bug fixes, and the ability to specify a custom http client for OpenAI AugmentedLLM (thanks @blevz!)

Changes

• Add ability to configure the http_client for openai @blevz (#173)

🐛 Bug Fixes

• fix: add system prompt handling to AzureAugmentedLLM @StreetLamb (#148)
• fx: _reasoning flag used to set api params @Pythonista7 (#139)
• Fixing spelling mistake in MCP Browser Agent README @andrew-lastmile (#175)
• Fixing broken links in the README @andrew-lastmile (#167)

Quality of Life

• Updated restructuring of examples directory @andrew-lastmile (#166)
• Restructuring the examples directory @andrew-lastmile (#144)
• SSE connection example and additional fixes for websockets @saqadri (#143)
• MCP-Agent: MCP Puppeteer to enable Console based browsing @andrew-lastmile (#137)
• feat: mcp_playwright_agent example - Updated @Haniehz1 (#156)

v0.0.16

Minor release adding proper support for MCP servers using SSE and Websockets, with configurable headers, as well as OpenAI o4 model support.

Huge thanks to @pmohan6 especially for adding WS support (with tests), as well as @aloktiagi for quickly adding o4 day-of-launch
 

Changes

• feature: support websocket connections and passing in HTTP headers to websockets and SSE connections @pmohan6 (#116)
• add support for o4 reasoning models @aloktiagi (#142)

v0.0.15

Highest number of contributors for this release! Thank you to all the wonderful open-source contributors for helping make mcp-agent better. Special shoutout to @StreetLamb for adding many of the key features in this release, includuing Azure OpenAI, AWS Bedrock, and Google Gemini model support!

Highlights

• AWS Bedrock support @StreetLamb (#85)
• Google Gemini support @StreetLamb (#129)
• Azure OpenAI support @StreetLamb (#90)
• Support mcp v1.6.0 (latest)

Coming up next

We are about to do a significant upgrade to mcp-agent in the next week. If you'd like to help contribute, here are our priorities:

• Add support for Temporal workflow executor for async agent orchestration (#14)
• Distributed tracing & observability (#10)
• Advanced MCP features (resources, prompts, sampling, and memory)
   

Changes

• Fix tool calling issue when using AWS Bedrock Nova models @StreetLamb (#134)
• Add support for Gemini models @StreetLamb (#129)
• fix: correct message handling in AzureAugmentedLLM to properly extend messages list @StreetLamb (#125)
• Updating the Github to Slack Agent with Github MCP Server @andrew-lastmile (#122)
• Fix OpenAI ChatCompletion message content format issue @askme765cs (#106)
• Fix: Functions Passed to Agents Are Now Available as Tools @rholinshead (#120)
• fix: refactor assistant message arguments in OpenAIAugmentedLLM @softage0 (#117)
• Update mcp_basic_azure_agent README.md @lucinvitae (#119)
• fix: move boto3-stubs to development dependencies @StreetLamb (#110)
• fix: use TYPE_CHECKING for conditional imports in augmented_llm_bedrock @StreetLamb (#109)
• Fix: Improve final response generation when reaching max iterations in AnthropicAugmentedLLM @dragon1086 (#99)
• fix: update tool result content formatting in AnthropicAugmentedLLM @StreetLamb (#95)
• update settings fetch to prioritize config_path config AND secrets @jtanningbed (#86)
• Add support for AWS profile in Bedrock settings @StreetLamb (#102)
• Adding example for a Github to Slack agent @andrew-lastmile (#98)
• Add Azure support @StreetLamb (#90)
• fix examples @boxbag (#92)
• fix a validation error when sending request to openai with name parameter @boxbag (#88)
• Add AWS Bedrock support @StreetLamb (#85)
• Fix duplicated system message (prompt cache) @recallfuture (#89)
• Hotfix for mcp > 1.5.0, and various updates from fast-agent repo @saqadri (#84)

🐛 Bug Fixes

• Fix generate_structured output in AzureAugmentedLLM and from_mcp_message_param issues @StreetLamb (#131)
• fix: correct extend messages logic in BedrockAugmentedLLM @StreetLamb (#130)

v0.0.11

Small qol updates

Changes

• Update instructor version to 1.7.7 to fix 567-labs/instructor#1412
• Fix structure and code block issue in README @skyduy (#71)
• readme: fix swarm agent example @sayanarijit (#70)
• Minor updates to examples

v0.0.10

Changes

• Improve server name parsing in tool calls @Henry-Jessie (#65)
• Add session_id to MCPApp, and propagate it through the system @corroleaus (#63)

Thank you to the community for continuing to help build this project! There's a major version update coming next.

v0.0.9

QoL updates over various examples, bug fixes, and a feature by @MattMorgis to allow multiple loggers to be configured simultaneously.
Thank you to all the excellent contributions from the community!

Changes

• fix example streamlit_mcp_rag_agent with last mcp-server-qdrant @venticello (#59)
• Relax numpy dependency constrain, now requiring numpy>=2.1.3 @Nicba1010 (#52)
• Logging Update @MattMorgis (#49)
• Multi transport logging @MattMorgis (#42)
• Fix bug arising from instantiation of CallToolResult class @meerkat-citronella (#48)

v0.0.8

Dedicating this release to @evalstate, who drove some incredible updates to the core framework, as well as some phenomenal examples, singlehandedly. Thank you! Details below:

Changes

• Add MCP roots support to mcp-agent @evalstate (#18). See mcp_root_test example for more details.
• Addition of ProgressLogger which captures the log event stream to convert log events to progress info @evalstate (#18).
• Intermediate progress reporting for visibility into agent/tool progress @evalstate (#18)
• Get Anthropic default model from config if available @MattMorgis (#39)
• Store context globally on init @andersenthomas98 (#37)
• fix finder_agent.py readme example @roterski (#28)
• Progress Reporting Feature @evalstate (#18)

Thank you to all the wonderful community contributors for making this release possible.

v0.0.7

Changes

• Ensure base_url is always used for OpenAI clients @mcolyer (#26)
• Fix duplicate tool call messages when using OpenAIAugmentedLLM @StreetLamb (#22)
• Support models compatible with OpenAI format @StreetLamb (#21)
• Add MCP agent server example @StreetLamb (#7)
• Fix NameError in streamlit_mcp_rag_agent example @StreetLamb (#9)
• Add Streamlit MCP RAG Agent example @StreetLamb (#4)
• example: Update marimo notebook README @akshayka (#6)
• example: add marimo notebook (basic mcp agent) @akshayka (#5)
• Add Streamlit MCP Agent example with configuration and requirements @StreetLamb (#3)
• Fix/secrets tracking @evalstate (#2)
• First commit: Create initial scaffolding for mcp-agent @saqadri (#1)

Thank you to all the wonderful community contributors for making this release possible.