Feat/extend llm apis

.env.example

@@ -4,19 +4,48 @@
 # starts.
 
 # ----------------------------------------------------------------------
-# Gemini (pick ONE of the two paths below)
+# LLM provider selection
 # ----------------------------------------------------------------------
+#
+# The server talks to an LLM through ``gee_mcp.server.llm.init_llm_client``,
+# which picks a provider from these two variables. Both are required for the
+# code-generation / analysis tools to work.
+#
+#   LLM_PROVIDER : one of ``google`` | ``anthropic`` | ``openai``
+#   LLM_NAME     : the model id for that provider, e.g.
+#                    google     -> gemini-3.1-pro-preview
+#                    anthropic  -> claude-opus-4-7
+#                    openai     -> gpt-5  (or an o-series reasoning model)
+LLM_PROVIDER=google
+LLM_NAME=gemini-3.1-pro-preview
+
+# Then set credentials for whichever provider you chose:
 
+# --- google: pick ONE of the two paths below ---
 # Path A: Gemini Developer API key.
 # GEMINI_API_KEY=
 # (``GOOGLE_API_KEY`` is also accepted as a fallback for
 # compatibility with Google's official SDK convention.)
-
+#
 # Path B — Vertex AI project (requires ``gcloud auth
 # application-default login`` to have been run)
 # VERTEXAI_PROJECT=your-vertexai-project
 # VERTEXAI_LOCATION=global
 
+# --- anthropic ---
+# ANTHROPIC_API_KEY=
+
+# --- openai ---
+# OPENAI_API_KEY=
+
+# ----------------------------------------------------------------------
+# LLM response cache (optional)
+# ----------------------------------------------------------------------
+#
+# ``init_llm_client(cache_dir=...)`` enables an on-disk JSON cache of LLM
+# responses (keyed by model + prompt). There is no env var for it yet —
+# pass ``cache_dir`` from code if you want it.
+
 # ----------------------------------------------------------------------
 # Google Earth Engine
 # ----------------------------------------------------------------------

README.md

@@ -119,30 +119,52 @@ Supported Python versions: 3.11–3.14.
 
 ## Configuration
 
-You need to configure access to **Gemini** and to **Google Earth
+You need to configure access to an **LLM provider** and to **Google Earth
 Engine** via environment variables. Copy [`.env.example`](.env.example)
 to `.env` and fill in the values; `python-dotenv` is loaded on server
 startup.
 
-### Gemini
+### LLM provider
 
-Either set an API key:
+The server talks to an LLM through a small pluggable layer
+(`gee_mcp.server.llm`). Pick a provider and model:
 
 | Variable | Required | Purpose |
 | --- | --- | --- |
-| `GEMINI_API_KEY` | yes | Gemini API key |
+| `LLM_PROVIDER` | yes | One of `google`, `anthropic`, `openai` |
+| `LLM_NAME` | yes | Model id for that provider (e.g. `gemini-3.1-pro-preview`, `claude-opus-4-7`, `gpt-5`) |
+
+Then set the credentials for the provider you chose:
+
+**`google`** — either an API key:
+
+| Variable | Required | Purpose |
+| --- | --- | --- |
+| `GEMINI_API_KEY` | yes | Gemini Developer API key |
 
 `GOOGLE_API_KEY` is also accepted as a fallback for compatibility
 with Google's official SDK convention.
 
-…or use a Vertex AI project (after running `gcloud auth
+…or a Vertex AI project (after running `gcloud auth
 application-default login`):
 
 | Variable | Required | Default | Purpose |
 | --- | --- | --- | --- |
 | `VERTEXAI_PROJECT` | yes |  | GCP project ID for Vertex AI |
 | `VERTEXAI_LOCATION` | no | `global` | GCP region for Vertex AI |
 
+**`anthropic`**:
+
+| Variable | Required | Purpose |
+| --- | --- | --- |
+| `ANTHROPIC_API_KEY` | yes | Anthropic API key |
+
+**`openai`**:
+
+| Variable | Required | Purpose |
+| --- | --- | --- |
+| `OPENAI_API_KEY` | yes | OpenAI API key |
+
 ### Google Earth Engine
 
 | Variable | Required | Purpose |