<html aria-label="message body"><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><div><br></div><div> I have (with lots of Satish’s and Hong’s help) stood up three PETSc related MCP servers, see <a href="https://urldefense.us/v3/__https://gitlab.com/petsc/petsc_mcp_servers/-/blob/main/README.md?ref_type=heads__;!!G_uCfscf7eWS!dbIU4Ye_9lCgMD3NAMbHwugpM6D0W21eir8f1I2_rmT2TlzU_nfglEf9uOzBAnzEdu5RHDHC280RJrcKti6cIg$">https://gitlab.com/petsc/petsc_mcp_servers/-/blob/main/README.md?ref_type=heads</a> for details on their usage. I hope anyone using AI/LLM tools in their use of PETSc can try out these servers and see if they help (or hinder) the use of AI development systems.</div><div><br></div><div> If you know of any available MCP servers that might help with PETSc please let us know. </div><div><br></div><div> Any problems using these or suggestions on making them better let us know.</div><div><br></div><div> Barry</div><div><br></div><div>Note that MCP servers/agents serve multiple purposes</div><div><br></div><div>- are used directly by developer systems like gemini-cli</div><div>- can be used directly by other agents we wrote</div><div>- can be used by modern RESTful interfaces for LLM based on the OpenAI response model</div><div><br></div><div>MCP servers can be used in multiple ways</div><div><br></div><div>- “directly” by an LLM, that is a LLM is given some text and a list of MCP tools and it decides exactly what tool to call with what arguments and</div><div>- via the MCP Python API; in this case it is just like programming Python and calling functions (that are transparently) actually implemented somewhere else.</div><div><br></div><div><br></div><div><br></div><div><br></div><div><br></div><div><br></div><div></div></body></html>