consult-llm-mcp 2.7.4 → 2.9.0

package/CHANGELOG.md

@@ -1,5 +1,24 @@
 # Changelog
 
+## v2.8.0 (2026-03-13)
+
+- Replaced hardcoded model enum with abstract selectors (`gemini`, `openai`,
+  `deepseek`) that resolve to the best available model at query time. This
+  avoids the need to hardcode a specific model in the caller side.
+- Responses now include a `[model:xxx]` prefix showing which concrete model was
+  used
+- Default Codex reasoning effort to "high" (was previously unset)
+- Monitor: added Task column to active and history tables
+- Monitor: show task mode and reasoning effort in detail view header
+- Monitor: press `s` in detail view to toggle system prompt display
+- Monitor: system prompt is now recorded in sidecar event files for viewing in
+  the TUI
+
+## v2.7.4 (2026-03-13)
+
+- Fixed Linux prebuilt binaries failing on older distros due to glibc version
+  mismatch by switching to musl static linking
+
 ## v2.7.1 (2026-03-09)
 
 - Monitor: show "Thinking..." spinner when thinking events are streaming

package/README.md

@@ -29,12 +29,10 @@ to bring in the heavy artillery. Supports multi-turn conversations.
 
 - Query powerful AI models (GPT-5.4, Gemini 3.1 Pro, DeepSeek Reasoner) with
   relevant files as context
-- Direct queries with optional file context
-- Include git changes for code review and analysis
-- Comprehensive logging with cost estimation
+- Include git changes for code review
+- Comprehensive logging with cost estimation (if using API)
 - [Monitor TUI](#monitor): Real-time dashboard for watching active consultations
-- [Gemini CLI backend](#gemini-cli): Use the `gemini` CLI to take advantage of
-  [free quota](https://developers.google.com/gemini-code-assist/resources/quotas#quotas-for-agent-mode-gemini-cli)
+- [Gemini CLI backend](#gemini-cli): Use the `gemini` CLI for Gemini models
 - [Codex CLI backend](#codex-cli): Use the `codex` CLI for OpenAI models
 - [Cursor CLI backend](#cursor-cli): Use the `cursor-agent` CLI to route GPT and
   Gemini models through a single tool
@@ -55,37 +53,40 @@ to bring in the heavy artillery. Supports multi-turn conversations.
 
    ```bash
    claude mcp add consult-llm \
-     -e OPENAI_API_KEY=your_key \
-     -e GEMINI_API_KEY=your_key \
+     -e CONSULT_LLM_GEMINI_BACKEND=gemini-cli \
+     -e CONSULT_LLM_OPENAI_BACKEND=codex-cli \
      -- npx -y consult-llm-mcp
    ```
 
+   This is the recommended setup. Uses [Gemini CLI](#gemini-cli) and
+   [Codex CLI](#codex-cli). No API keys required, just `gemini login` and
+   `codex login`.
+
    **With binary** (no Node.js required):
 
    ```bash
    curl -fsSL https://raw.githubusercontent.com/raine/consult-llm-mcp/main/scripts/install.sh | bash
+   ```
+
+   ```bash
    claude mcp add consult-llm \
-     -e OPENAI_API_KEY=your_key \
-     -e GEMINI_API_KEY=your_key \
+     -e CONSULT_LLM_GEMINI_BACKEND=gemini-cli \
+     -e CONSULT_LLM_OPENAI_BACKEND=codex-cli \
      -- consult-llm-mcp
    ```
 
    For global availability across projects, add `--scope user`.
 
-   <details>
-   <summary>Using multiple API keys or CLI backends</summary>
+   **Using API keys instead of CLI backends:**
 
    ```bash
    claude mcp add consult-llm \
      -e OPENAI_API_KEY=your_openai_key \
      -e GEMINI_API_KEY=your_gemini_key \
      -e DEEPSEEK_API_KEY=your_deepseek_key \
-     -e CONSULT_LLM_GEMINI_BACKEND=gemini-cli \
      -- npx -y consult-llm-mcp
    ```
 
-   </details>
-
 2. **Verify connection** with `/mcp`:
 
    ```
@@ -380,8 +381,9 @@ claude mcp add consult-llm -e CONSULT_LLM_OPENAI_BACKEND=codex-cli -- npx -y con
 
 <!-- prettier-ignore -->
 > [!TIP]
-> Set reasoning effort with `-e CONSULT_LLM_CODEX_REASONING_EFFORT=high`. Options:
-> `none`, `minimal`, `low`, `medium`, `high`, `xhigh`.
+> Reasoning effort defaults to `high`. Override with
+> `-e CONSULT_LLM_CODEX_REASONING_EFFORT=xhigh`. Options: `none`, `minimal`,
+> `low`, `medium`, `high`, `xhigh`.
 
 #### Cursor CLI
 
@@ -405,8 +407,7 @@ claude mcp add consult-llm -e CONSULT_LLM_GEMINI_BACKEND=cursor-cli -- npx -y co
 claude mcp add consult-llm \
   -e CONSULT_LLM_OPENAI_BACKEND=cursor-cli \
   -e CONSULT_LLM_GEMINI_BACKEND=cursor-cli \
-  -e CONSULT_LLM_CODEX_REASONING_EFFORT=high \
-  -e CONSULT_LLM_ALLOWED_MODELS="gemini-3-pro-preview,gpt-5.3-codex" \
+  -e CONSULT_LLM_ALLOWED_MODELS="gemini-3.1-pro-preview,gpt-5.3-codex" \
   -- npx -y consult-llm-mcp
 ```
 
@@ -419,11 +420,7 @@ review), allow them in `~/.cursor/cli-config.json`:
 ```json
 {
   "permissions": {
-    "allow": [
-      "Shell(git diff*)",
-      "Shell(git log*)",
-      "Shell(git show*)"
-    ],
+    "allow": ["Shell(git diff*)", "Shell(git log*)", "Shell(git show*)"],
     "deny": []
   }
 }
@@ -495,31 +492,38 @@ See the "Using web mode..." example above for a concrete transcript.
   mode)
 - `DEEPSEEK_API_KEY` - Your DeepSeek API key (required for DeepSeek models)
 - `CONSULT_LLM_DEFAULT_MODEL` - Override the default model (optional)
-  - Options: `gpt-5.2` (default), `gpt-5.4`, `gemini-2.5-pro`,
-    `gemini-3-pro-preview`, `gemini-3.1-pro-preview`, `deepseek-reasoner`,
-    `gpt-5.3-codex`, `gpt-5.2-codex`
+  - Accepts selectors (`gemini`, `openai`, `deepseek`) or exact model IDs
+    (`gpt-5.4`, `gemini-3.1-pro-preview`, etc.)
+  - Selectors are resolved to the best available model at startup
 - `CONSULT_LLM_GEMINI_BACKEND` - Backend for Gemini models (optional)
   - Options: `api` (default), `gemini-cli`, `cursor-cli`
 - `CONSULT_LLM_OPENAI_BACKEND` - Backend for OpenAI models (optional)
   - Options: `api` (default), `codex-cli`, `cursor-cli`
-- `CONSULT_LLM_CODEX_REASONING_EFFORT` - Configure reasoning effort for Codex
-  CLI (optional)
-  - See [Codex CLI](#codex-cli) for details and available options
+- `CONSULT_LLM_ALLOWED_MODELS` - Restrict which concrete models can be used
+  (optional)
+  - Comma-separated list, e.g., `gpt-5.4,gemini-3.1-pro-preview`
+  - Selectors resolve against this list — e.g., if only `gemini-2.5-pro` is
+    allowed, the `gemini` selector resolves to it
+  - Useful when a backend doesn't support all models (e.g., Cursor CLI)
+  - See [Tips](#controlling-which-models-are-used) for usage examples
 - `CONSULT_LLM_EXTRA_MODELS` - Add models not in the built-in list (optional)
   - Comma-separated list, e.g., `grok-3,kimi-k2.5`
   - Merged with built-in models and included in the tool schema
   - Useful for newly released models with a known provider prefix (`gpt-`,
     `gemini-`, `deepseek-`)
-- `CONSULT_LLM_ALLOWED_MODELS` - List of models to advertise (optional)
-  - Comma-separated list, e.g., `gpt-5.2,gemini-3-pro-preview`
-  - When set, only these models appear in the tool schema
-  - Filters the combined catalog (built-in + extra models)
-  - If `CONSULT_LLM_DEFAULT_MODEL` is set, it must be in this list
-  - See [Tips](#controlling-which-models-claude-uses) for usage examples
+- `CONSULT_LLM_CODEX_REASONING_EFFORT` - Configure reasoning effort for Codex
+  CLI (optional, default: `high`)
+  - See [Codex CLI](#codex-cli) for details and available options
 - `CONSULT_LLM_SYSTEM_PROMPT_PATH` - Custom path to system prompt file
   (optional)
   - Overrides the default `~/.consult-llm-mcp/SYSTEM_PROMPT.md` location
   - Useful for project-specific prompts
+- `CONSULT_LLM_NO_UPDATE_CHECK` - Disable automatic update checking on server
+  startup (optional)
+  - Set to `1` to disable
+  - By default, the server checks for new versions in the background every 24
+    hours and logs a notice when an update is available
+  - Only applies to binary installs — npm installs are never checked
 - `MCP_DEBUG_STDIN` - Log raw JSON-RPC messages received on stdin (optional)
   - Set to `1` to enable
   - Logs every message as `RAW RECV` entries and poll timing gaps as
@@ -558,30 +562,33 @@ claude mcp add consult-llm \
 
 ## Tips
 
-### Controlling which models Claude uses
+### Controlling which models are used
 
-When you ask Claude to "consult an LLM" without specifying a model, it picks one
-from the available options in the tool schema. The `CONSULT_LLM_DEFAULT_MODEL`
-only affects the fallback when no model is specified in the tool call.
+The `model` parameter accepts **selectors** (`gemini`, `openai`, `deepseek`)
+that the server resolves to the best available concrete model. When no model is
+specified, the server uses `CONSULT_LLM_DEFAULT_MODEL` or its built-in fallback.
 
-To control which models Claude can choose from, use
-`CONSULT_LLM_ALLOWED_MODELS`:
+**Selector resolution order** (first available wins):
 
-```bash
-claude mcp add consult-llm \
-  -e GEMINI_API_KEY=your_key \
-  -e CONSULT_LLM_ALLOWED_MODELS='gemini-3-pro-preview,gpt-5.2-codex' \
-  -- npx -y consult-llm-mcp
-```
+| Selector   | Priority                                                       |
+| ---------- | -------------------------------------------------------------- |
+| `gemini`   | gemini-3.1-pro-preview → gemini-3-pro-preview → gemini-2.5-pro |
+| `openai`   | gpt-5.4 → gpt-5.3-codex → gpt-5.2 → gpt-5.2-codex              |
+| `deepseek` | deepseek-reasoner                                              |
 
-This restricts the tool schema to only advertise these models. For example, to
-ensure Claude always uses Gemini 3 Pro:
+**Restricting models with `CONSULT_LLM_ALLOWED_MODELS`:**
+
+If your backend doesn't support all models (e.g., Cursor CLI can't use
+`gpt-5.4`), use `CONSULT_LLM_ALLOWED_MODELS` to filter. Selectors will
+automatically resolve to the best model within the allowed list:
 
 ```bash
+# Only allow codex models through Cursor CLI
 claude mcp add consult-llm \
-  -e GEMINI_API_KEY=your_key \
-  -e CONSULT_LLM_ALLOWED_MODELS='gemini-3-pro-preview' \
+  -e CONSULT_LLM_OPENAI_BACKEND=cursor-cli \
+  -e CONSULT_LLM_ALLOWED_MODELS='gpt-5.3-codex,gemini-3.1-pro-preview' \
   -- npx -y consult-llm-mcp
+# "openai" selector → gpt-5.3-codex (gpt-5.4 filtered out)
 ```
 
 ## MCP tool: consult_llm
@@ -596,10 +603,12 @@ models complex questions.
 - **files** (optional): Array of file paths to include as context
   - All files are added as context with file paths and code blocks
 
-- **model** (optional): LLM model to use
-  - Options: `gpt-5.2` (default), `gpt-5.4`, `gemini-2.5-pro`,
-    `gemini-3-pro-preview`, `gemini-3.1-pro-preview`, `deepseek-reasoner`,
-    `gpt-5.3-codex`, `gpt-5.2-codex`
+- **model** (optional): Model selector or exact model ID
+  - Selectors: `gemini`, `openai`, `deepseek` — the server resolves to the best
+    available model for each family
+  - Exact model IDs (`gpt-5.4`, `gemini-3.1-pro-preview`, etc.) are also
+    accepted as an advanced override
+  - When omitted, the server uses the configured default
 
 - **task_mode** (optional): Controls the system prompt persona. The calling LLM
   should choose based on the task:
@@ -631,15 +640,12 @@ models complex questions.
 
 ## Supported models
 
-- **gemini-2.5-pro**: Google's Gemini 2.5 Pro ($1.25/$10 per million tokens)
-- **gemini-3-pro-preview**: Google's Gemini 3 Pro Preview ($2/$12 per million
-  tokens for prompts ≤200k tokens, $4/$18 for prompts >200k tokens)
-- **gemini-3.1-pro-preview**: Google's Gemini 3.1 Pro Preview ($2/$12 per
-  million tokens for prompts ≤200k tokens, $4/$18 for prompts >200k tokens)
-- **deepseek-reasoner**: DeepSeek's reasoning model ($0.55/$2.19 per million
-  tokens)
-- **gpt-5.4**: OpenAI's GPT-5.4 model ($2.50/$15 per million tokens)
-- **gpt-5.2**: OpenAI's GPT-5.2 model ($1.75/$14 per million tokens)
+- **gemini-2.5-pro**: Google's Gemini 2.5 Pro
+- **gemini-3-pro-preview**: Google's Gemini 3 Pro Preview
+- **gemini-3.1-pro-preview**: Google's Gemini 3.1 Pro Preview
+- **deepseek-reasoner**: DeepSeek's reasoning model
+- **gpt-5.4**: OpenAI's GPT-5.4 model
+- **gpt-5.2**: OpenAI's GPT-5.2 model
 - **gpt-5.3-codex**: OpenAI's Codex model based on GPT-5.3
 - **gpt-5.2-codex**: OpenAI's Codex model based on GPT-5.2
 
@@ -739,7 +745,9 @@ always reliably triggered. See the [consult skill](#consult) below.
 **Recommendation:** Start with no custom activation. Use skills if you need
 custom instructions for how the MCP is invoked.
 
-## Installing skills
+## Skills
+
+### Installing skills
 
 Install all skills globally with a single command:
 
@@ -759,8 +767,6 @@ To uninstall:
 curl -fsSL https://raw.githubusercontent.com/raine/consult-llm-mcp/main/scripts/install-skills | bash -s uninstall
 ```
 
-## Skills
-
 ### consult
 
 An example [Claude Code skill](https://code.claude.com/docs/en/skills) that uses
@@ -816,6 +822,21 @@ forth before synthesizing and implementing. See
 > /debate-vs --gemini design the multi-tenant isolation strategy
 ```
 
+## Updating
+
+**Binary installs:**
+
+```bash
+consult-llm-mcp update
+```
+
+Downloads the latest release from GitHub with SHA-256 checksum verification. If
+`consult-llm-monitor` is found alongside the binary, it's updated too.
+
+The server also checks for updates in the background on startup (every 24 hours)
+and logs a notice when a newer version is available. Disable with
+`CONSULT_LLM_NO_UPDATE_CHECK=1`.
+
 ## Development
 
 To work on the MCP server locally and use your development version:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "consult-llm-mcp",
-  "version": "2.7.4",
+  "version": "2.9.0",
   "description": "MCP server for consulting powerful AI models",
   "repository": {
     "type": "git",
@@ -31,9 +31,9 @@
     "ai"
   ],
   "optionalDependencies": {
-    "consult-llm-mcp-darwin-arm64": "2.7.4",
-    "consult-llm-mcp-darwin-x64": "2.7.4",
-    "consult-llm-mcp-linux-x64": "2.7.4",
-    "consult-llm-mcp-linux-arm64": "2.7.4"
+    "consult-llm-mcp-darwin-arm64": "2.9.0",
+    "consult-llm-mcp-darwin-x64": "2.9.0",
+    "consult-llm-mcp-linux-x64": "2.9.0",
+    "consult-llm-mcp-linux-arm64": "2.9.0"
   }
 }