Add keyless Parallel search for unconfigured issue analysis

README.md

@@ -271,6 +271,25 @@ Verify at [http://localhost:7474](http://localhost:7474)
 
 ---
 
+### Web search
+
+Issue-analysis agents use [Parallel Search MCP](https://docs.parallel.ai/integrations/mcp/search-mcp)
+when no Tavily key is configured. Search runs in Fast mode without an account or API key.
+The anonymous service has usage limits; requests can return quota errors.
+
+`PROMETHEUS_WEB_SEARCH_PROVIDER` defaults to `auto`: an existing
+`PROMETHEUS_TAVILY_API_KEY` keeps Tavily selected. Set the provider to `parallel`
+or `tavily` to choose explicitly, or `disabled` to remove web search from the agents.
+An explicitly selected Tavily provider requires a Tavily key. Provider failures
+are reported without switching providers. `PROMETHEUS_WEB_SEARCH_TIMEOUT` bounds
+Parallel requests, including connection setup (30 seconds by default).
+
+Parallel uses server-managed result counts and source selection. Custom result
+limits and domain filters remain available through Tavily; requesting them with
+Parallel raises an error. The agent-facing tool accepts a `query` with either provider.
+Queries are sent to the selected external service; use `disabled` when external
+web search is not permitted.
+
 ## 🤝 Contributing
 
 We welcome contributions! Please see our **[Contributing Guidelines](CONTRIBUTING.md)** for details on how to get started.

example.env

@@ -33,8 +33,12 @@ PROMETHEUS_ADVANCED_MODEL_TEMPERATURE=0.5
 
 PROMETHEUS_BASE_MODEL_TEMPERATURE=0.5
 
-# Tavily API settings (You can sign up for a free API key at https://www.tavily.com/)
-PROMETHEUS_TAVILY_API_KEY=your_tavily_api_key
+# Web search: auto uses a configured Tavily key, otherwise free Parallel Search MCP.
+# Use parallel or tavily to select explicitly, or disabled to remove search tools.
+PROMETHEUS_WEB_SEARCH_PROVIDER=auto
+PROMETHEUS_WEB_SEARCH_TIMEOUT=30
+# Optional Tavily API key (https://www.tavily.com/)
+# PROMETHEUS_TAVILY_API_KEY=your_tavily_api_key
 
 # Database settings
 PROMETHEUS_DATABASE_URL=postgresql+asyncpg://postgres:password@postgres:5432/postgres

prometheus/configuration/config.py

@@ -1,5 +1,6 @@
 from typing import List, Literal, Optional
 
+from pydantic import Field
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
 
@@ -62,8 +63,10 @@ class Settings(BaseSettings):
     # Default normal user repository number
     DEFAULT_USER_REPOSITORY_LIMIT: int = 5
 
-    # tool for Websearch
-    TAVILY_API_KEY: str
+    # Web search: retain configured Tavily access; otherwise use keyless Parallel.
+    WEB_SEARCH_PROVIDER: Literal["auto", "parallel", "tavily", "disabled"] = "auto"
+    WEB_SEARCH_TIMEOUT: float = Field(default=30.0, gt=0, le=300)
+    TAVILY_API_KEY: Optional[str] = None
 
     # Athena semantic memory service
     ATHENA_BASE_URL: Optional[str] = None

prometheus/lang_graph/nodes/issue_bug_analyzer_node.py

@@ -49,12 +49,7 @@ class IssueBugAnalyzerNode:
 
 Tools available:
 - web_search: Searches the web for technical information to aid in bug analysis and resolution.
-When using the web_search tool, it's recommended to include these parameters:
-    - include_domains: ['stackoverflow.com', 'github.com', 'developer.mozilla.org', 'learn.microsoft.com', 'fastapi.tiangolo.com'
-            'docs.python.org', 'pydantic.dev', 'pypi.org', 'readthedocs.org', 'docs.djangoproject.com','flask.palletsprojects.com']
-    - search_depth: "advanced"
-
-    Make sure to explicitly pass these parameters in your tool call.
+Pass a specific technical query to web_search.
 
 Important:
 - Do NOT provide actual code snippets or diffs
@@ -72,16 +67,19 @@ def __init__(self, model: BaseChatModel):
         self.model = model
         self.system_prompt = SystemMessage(self.SYS_PROMPT)
         self.tools = self._init_tools()
-        self.model_with_tools = model.bind_tools(self.tools)
+        self.model_with_tools = model.bind_tools(self.tools) if self.tools else model
         self._logger = logging.getLogger(f"thread-{threading.get_ident()}.{__name__}")
 
     def _init_tools(self):
         """Initializes tools for the node."""
         tools = []
+        if self.web_search_tool.provider == "disabled":
+            return tools
 
         web_search_fn = functools.partial(self.web_search_tool.web_search)
         web_search_tool = StructuredTool.from_function(
             func=web_search_fn,
+            coroutine=self.web_search_tool.aweb_search,
             name=self.web_search_tool.web_search.__name__,
             description=self.web_search_tool.web_search_spec.description,
             args_schema=self.web_search_tool.web_search_spec.input_schema,

prometheus/lang_graph/nodes/issue_documentation_analyzer_node.py

@@ -44,17 +44,20 @@ def __init__(self, model: BaseChatModel):
         self.system_prompt = SystemMessage(self.SYS_PROMPT)
         self.web_search_tool = WebSearchTool()
         self.tools = self._init_tools()
-        self.model_with_tools = model.bind_tools(self.tools)
+        self.model_with_tools = model.bind_tools(self.tools) if self.tools else model
 
         self._logger = logging.getLogger(f"thread-{threading.get_ident()}.{__name__}")
 
     def _init_tools(self):
         """Initializes tools for the node."""
         tools = []
+        if self.web_search_tool.provider == "disabled":
+            return tools
 
         web_search_fn = functools.partial(self.web_search_tool.web_search)
         web_search_tool = StructuredTool.from_function(
             func=web_search_fn,
+            coroutine=self.web_search_tool.aweb_search,
             name=self.web_search_tool.web_search.__name__,
             description=self.web_search_tool.web_search_spec.description,
             args_schema=self.web_search_tool.web_search_spec.input_schema,

prometheus/lang_graph/nodes/issue_feature_analyzer_node.py

@@ -50,12 +50,7 @@ class IssueFeatureAnalyzerNode:
 
 Tools available:
 - web_search: Searches the web for technical information to aid in feature design and implementation.
-When using the web_search tool, it's recommended to include these parameters:
-    - include_domains: ['stackoverflow.com', 'github.com', 'developer.mozilla.org', 'learn.microsoft.com', 'fastapi.tiangolo.com'
-            'docs.python.org', 'pydantic.dev', 'pypi.org', 'readthedocs.org', 'docs.djangoproject.com','flask.palletsprojects.com']
-    - search_depth: "advanced"
-
-    Make sure to explicitly pass these parameters in your tool call.
+Pass a specific technical query to web_search.
 
 Important:
 - Do NOT provide actual code snippets or diffs
@@ -74,16 +69,19 @@ def __init__(self, model: BaseChatModel):
         self.model = model
         self.system_prompt = SystemMessage(self.SYS_PROMPT)
         self.tools = self._init_tools()
-        self.model_with_tools = model.bind_tools(self.tools)
+        self.model_with_tools = model.bind_tools(self.tools) if self.tools else model
         self._logger = logging.getLogger(f"thread-{threading.get_ident()}.{__name__}")
 
     def _init_tools(self):
         """Initializes tools for the node."""
         tools = []
+        if self.web_search_tool.provider == "disabled":
+            return tools
 
         web_search_fn = functools.partial(self.web_search_tool.web_search)
         web_search_tool = StructuredTool.from_function(
             func=web_search_fn,
+            coroutine=self.web_search_tool.aweb_search,
             name=self.web_search_tool.web_search.__name__,
             description=self.web_search_tool.web_search_spec.description,
             args_schema=self.web_search_tool.web_search_spec.input_schema,

prometheus/lang_graph/nodes/issue_question_analyzer_node.py

@@ -86,17 +86,20 @@ def __init__(self, model: BaseChatModel):
         self.system_prompt = SystemMessage(self.SYS_PROMPT)
         self.web_search_tool = WebSearchTool()
         self.tools = self._init_tools()
-        self.model_with_tools = model.bind_tools(self.tools)
+        self.model_with_tools = model.bind_tools(self.tools) if self.tools else model
 
         self._logger = logging.getLogger(f"thread-{threading.get_ident()}.{__name__}")
 
     def _init_tools(self):
         """Initializes tools for the node."""
         tools = []
+        if self.web_search_tool.provider == "disabled":
+            return tools
 
         web_search_fn = functools.partial(self.web_search_tool.web_search)
         web_search_tool = StructuredTool.from_function(
             func=web_search_fn,
+            coroutine=self.web_search_tool.aweb_search,
             name=self.web_search_tool.web_search.__name__,
             description=self.web_search_tool.web_search_spec.description,
             args_schema=self.web_search_tool.web_search_spec.input_schema,

prometheus/tools/web_search.py

@@ -1,8 +1,12 @@
+import asyncio
 import logging
 import threading
 from dataclasses import dataclass
 from typing import Annotated
 
+from mcp import ClientSession
+from mcp.client.streamable_http import streamablehttp_client
+from mcp.types import Implementation
 from pydantic import BaseModel, Field
 from tavily import InvalidAPIKeyError, TavilyClient, UsageLimitExceededError
 
@@ -17,7 +21,7 @@ class ToolSpec:
 
 
 class WebSearchInput(BaseModel):
-    """Base parameters for Tavily search."""
+    """Parameters shared by the web search providers."""
 
     query: Annotated[str, Field(description="Search query")]
 
@@ -79,8 +83,12 @@ def __init__(self):
         self._logger = logging.getLogger(f"thread-{threading.get_ident()}.{__name__}")
 
         tavily_api_key = settings.TAVILY_API_KEY
-        if tavily_api_key is None:
-            self._logger.warning("Tavily API key is not set")
+        provider = settings.WEB_SEARCH_PROVIDER
+        self.provider = (
+            ("tavily" if tavily_api_key else "parallel") if provider == "auto" else provider
+        )
+        self.timeout = settings.WEB_SEARCH_TIMEOUT
+        if self.provider != "tavily" or not tavily_api_key:
             tavily_client = None
         else:
             tavily_client = TavilyClient(api_key=tavily_api_key)
@@ -89,21 +97,30 @@ def __init__(self):
     def web_search(
         self,
         query: str,
-        max_results: int = 5,
+        max_results: int | None = None,
         include_domains=None,
-        exclude_domains: list[str] = None,
+        exclude_domains: list[str] | None = None,
     ) -> str:
         """Search the web for technical information to aid in bug analysis and resolution.
 
         Args:
             query: Search query string.
-            max_results: Maximum number of results to return (default: 5).
-            include_domains: List of domains to include in search.
-            exclude_domains: List of domains to exclude from search.
+            max_results: Tavily result limit (default: 5); Parallel uses server settings.
+            include_domains: Tavily domains to include; unsupported on Parallel.
+            exclude_domains: Tavily domains to exclude; unsupported on Parallel.
 
         Returns:
             Formatted search results as a string.
         """
+        if self.provider == "parallel":
+            return asyncio.run(
+                self.aweb_search(query, max_results, include_domains, exclude_domains)
+            )
+        if self.provider == "disabled":
+            raise WebSearchToolException("Web search is disabled")
+        if self.tavily_client is None:
+            raise WebSearchToolException("Tavily requires PROMETHEUS_TAVILY_API_KEY")
+
         # Set default include domains if not provided
         if include_domains is None:
             include_domains = [
@@ -123,7 +140,7 @@ def web_search(
         try:
             response = self.tavily_client.search(
                 query=query,
-                max_results=max_results,
+                max_results=5 if max_results is None else max_results,
                 search_depth="advanced",
                 include_answer=True,
                 include_domains=include_domains or [],  # Convert None to an empty list
@@ -140,3 +157,55 @@ def web_search(
         format_response = format_results(response)
         self._logger.info(f"web_search format_response: {format_response}")
         return format_response
+
+    async def aweb_search(
+        self,
+        query: str,
+        max_results: int | None = None,
+        include_domains=None,
+        exclude_domains: list[str] | None = None,
+    ) -> str:
+        """Search asynchronously, allowing Parallel requests to be cancelled."""
+        if self.provider != "parallel":
+            return await asyncio.to_thread(
+                self.web_search, query, max_results, include_domains, exclude_domains
+            )
+        if max_results is not None or include_domains or exclude_domains:
+            raise WebSearchToolException(
+                "Parallel uses server-managed search settings. Select the tavily provider "
+                "to use domain filters or custom result limits."
+            )
+        try:
+            return await asyncio.wait_for(self._parallel_search(query), timeout=self.timeout)
+        except TimeoutError:
+            raise WebSearchToolException("Parallel search timed out") from None
+        except WebSearchToolException:
+            raise
+        except Exception:  # noqa: BLE001 - sanitize SDK exceptions before model consumption.
+            # SDK transport errors can contain request details. Keep the public error concise.
+            raise WebSearchToolException("Parallel Search MCP request failed") from None
+
+    async def _parallel_search(self, query: str) -> str:
+        # Each call owns and closes its transport. Do not attach a process/client-wide
+        # conversation ID: analyzer instances are reused across independent issue runs.
+        headers = {"User-Agent": "Prometheus/1.3 (+https://github.com/EuniAI/Prometheus)"}
+        async with (
+            streamablehttp_client("https://search.parallel.ai/mcp", headers=headers) as (
+                read,
+                write,
+                _,
+            ),
+            ClientSession(
+                read, write, client_info=Implementation(name="Prometheus", version="1.3")
+            ) as session,
+        ):
+            await session.initialize()
+            result = await session.call_tool(
+                "web_search", {"objective": query, "search_queries": [query]}
+            )
+        text = "\n".join(block.text for block in result.content if block.type == "text")
+        if result.isError:
+            raise WebSearchToolException(f"Parallel Search MCP returned an error: {text}")
+        if not text.strip():
+            raise WebSearchToolException("Parallel Search MCP returned no text results")
+        return text

pyproject.toml

@@ -27,7 +27,9 @@ dependencies = [
   "sqlmodel==0.0.24",
   "asyncpg",
   "pyjwt==2.6.0",
-  "mcp>=1.4.1",
+  "mcp>=1.10.1,<2",
+  # MCP versions compatible with the pinned PyJWT require Pydantic <2.12.
+  "pydantic>=2.7,<2.12",
   "tavily-python>=0.5.1",
   "langchain-mcp-adapters>=0.1.9",
   "httpx==0.28.1",