Describe the bug
The server crashes on startup and never completes MCP initialization. All tool groups register successfully, then main() calls cache.ensure_ready(), which fetches the hardcoded docs index at:
https://aws.github.io/bedrock-agentcore-starter-toolkit/llms.txt
That URL now returns HTTP 404. The startup fetch has no error handling, so theurllib.error.HTTPError escapes out of main() and the process exits. The MCP client sees the stdio server die and marks it as failed.
The root cause is external: the aws/bedrock-agentcore-starter-toolkit docs site has been deprecated, and the MkDocs llmstxt plugin that generated llms.txt was removed. From that repo's documentation/mkdocs.yaml:
SITE DEPRECATION: the content-generating plugins (macros, include-markdown, mkdocstrings, llmstxt) were removed. Every page now serves a redirect ...
The starter toolkit is heading to EOL in favor of the new AgentCore CLI, and this GitHub Pages site is being deprecated.
So llms.txt is gone permanently at that location; both llms.txt and llms-full.txt return 404.
Expected Behavior
A transient or permanent failure to fetch the configured llms.txt should not prevent the server from starting. Documentation search/fetch tools can degrade gracefully, and the unrelated tools should remain usable.
Current Behavior
Process exits with a traceback during startup:
INFO ... Code interpreter tools registered (9 tools)
Traceback (most recent call last):
File ".../bin/awslabs.amazon-bedrock-agentcore-mcp-server", line 12, in <module>
sys.exit(main())
File ".../server.py", line 261, in main
cache.ensure_ready()
File ".../utils/cache.py", line 74, in ensure_ready
load_links_only()
File ".../utils/cache.py", line 48, in load_links_only
for title, url in doc_fetcher.parse_llms_txt(src):
File ".../utils/doc_fetcher.py", line 77, in parse_llms_txt
txt = _get(url)
File ".../utils/doc_fetcher.py", line 60, in _get
with urllib.request.urlopen(req, timeout=doc_config.timeout) as r: # nosec
...
urllib.error.HTTPError: HTTP Error 404: Not Found
Reproduction Steps
Run uvx awslabs.amazon-bedrock-agentcore-mcp-server@latest
Possible Solution
No response
Additional Information/Context
No response
OS
platform-independent
Server
amazon-bedrock-agentcore-mcp-server
Server Version
@latest - and all prior versions
Region experiencing the issue
N/A
Other information
No response
Service quota
Describe the bug
The server crashes on startup and never completes MCP initialization. All tool groups register successfully, then
main()callscache.ensure_ready(), which fetches the hardcoded docs index at:https://aws.github.io/bedrock-agentcore-starter-toolkit/llms.txt
That URL now returns HTTP 404. The startup fetch has no error handling, so the
urllib.error.HTTPErrorescapes out ofmain()and the process exits. The MCP client sees the stdio server die and marks it as failed.The root cause is external: the aws/bedrock-agentcore-starter-toolkit docs site has been deprecated, and the MkDocs llmstxt plugin that generated llms.txt was removed. From that repo's documentation/mkdocs.yaml:
So llms.txt is gone permanently at that location; both llms.txt and llms-full.txt return 404.
Expected Behavior
A transient or permanent failure to fetch the configured llms.txt should not prevent the server from starting. Documentation search/fetch tools can degrade gracefully, and the unrelated tools should remain usable.
Current Behavior
Process exits with a traceback during startup:
Reproduction Steps
Run
uvx awslabs.amazon-bedrock-agentcore-mcp-server@latestPossible Solution
No response
Additional Information/Context
No response
OS
platform-independent
Server
amazon-bedrock-agentcore-mcp-server
Server Version
@latest - and all prior versions
Region experiencing the issue
N/A
Other information
No response
Service quota