<tool> <purpose>Finds relevant code in an indexed codebase using natural language or keyword queries</purpose> <when_to_use> <scenario>Find specific functions, classes, or code patterns</scenario> <scenario>Get context before making changes to understand dependencies</scenario> <scenario>Explore ...
Risk signalsAccepts freeform code/query input (query)
Part of the Deepcontext server.
Free to start. No card required.
AI agents call search_codebase to retrieve information from Deepcontext without modifying any data. This is common in research, monitoring, and reporting workflows where the agent needs context before taking action. Because read operations don't change state, they are generally safe to allow without restrictions -- but you may still want rate limits to control API costs.
Even though search_codebase only reads data, uncontrolled read access can leak sensitive information or rack up API costs. An agent caught in a retry loop could make thousands of calls per minute. A rate limit gives you a safety net without blocking legitimate use.
Read-only tools are safe to allow by default. No rate limit needed unless you want to control costs.
{
"version": "1",
"default": "deny",
"tools": {
"search_codebase": {}
}
}See the full Deepcontext policy for all 4 tools.
These attack patterns abuse exactly the kind of access search_codebase gives an agent. Each links to the full case and the policy that stops it:
Other read tools across the catalogue. The same approach applies to each: allow, with a rate cap to control cost.
<tool> <purpose>Finds relevant code in an indexed codebase using natural language or keyword queries</purpose> <when_to_use> <scenario>Find specific functions, classes, or code patterns</scenario> <scenario>Get context before making changes to understand dependencies</scenario> <scenario>Explore how existing systems work</scenario> <scenario>Locate examples of API usage or patterns</scenario> </when_to_use> <parameters> <parameter name="query" required="true"> <type>string</type> <description>Natural language or keyword search query describing what code to find</description> </parameter> <parameter name="codebase_path" required="true"> <type>string</type> <description>ABSOLUTE path to the codebase directory to search</description> <examples> <valid>/Users/name/project</valid> <valid>/home/user/code/repo</valid> <invalid>.</invalid> <invalid>../project</invalid> <invalid>relative/path</invalid> </examples> <validation>Must be absolute path starting with / (Unix) or C:\ (Windows)</validation> </parameter> <parameter name="max_results" required="false"> <type>number</type> <description>Maximum number of code chunks to return</description> <default>5</default> <best_practice>Keep at default 5 for focused results. Use multiple targeted searches rather than increasing this limit</best_practice> </parameter> </parameters> <strategy> <guideline>Use specific technical terms: "authentication middleware", "database connection", "error handler"</guideline> <guideline>Focus on implementation: "user login function" rather than "user management system"</guideline> <guideline>Include file types when relevant: "SQL migration", "React component", "API endpoint"</guideline> </strategy> <workflow> <step>Search discovers relevant files and entry points, use imports and exports to find related files</step> <step>Use Read tool to explore discovered files in detail for complete implementation</step> <step>Use Grep tool for precise pattern matching of specific symbols or exact text</step> <step>Follow imports/exports from results to guide next searches</step> <step>Prefer multiple focused searches with 5 results over single large searches</step> <step>Search provides discovery, not complete solutions</step> </workflow> <result_interpretation> <point>Results ranked by semantic relevance, not code importance</point> <point>Implementation code often appears in results 2-5, not just #1</point> <point>Look for actual code files (.ts, .js, .sql) over documentation (.md, .txt)</point> </result_interpretation> <limitations> <limitation> <description>May miss foundational type definitions</description> <solution>Use Grep for "interface PluginName"</solution> </limitation> <limitation> <description>Shows implementations, not core contracts</description> <solution>Follow up with Read for full context</solution> </limitation> <limitation> <description>Semantic chunks may lack architectural hierarchy</description> <solution>Manual file exploration needed</solution> </limitation> <limitation> <description>Excludes filtered content: test files, generated code, config files, minified files, large data files</description> <solution>Use Grep tool to search test files (*.test.*, *.spec.*, __tests__, /tests/), config files, or generated content</solution> </limitation> <limitation> <description>For precise symbol search</description> <solution>Use Grep tool for exact matches</solution> </limitation> </limitations> <returns>Code chunks with file paths, line numbers, relevance scores, symbol information, imports, and exports</returns> <prerequisites>Codebase must be indexed first with index_codebase</prerequisites> </tool>. It is categorised as a Read tool in the Deepcontext MCP Server, which means it retrieves data without modifying state.
Register the Deepcontext MCP server in PolicyLayer and add a rule for search_codebase: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Deepcontext. Nothing to install.
search_codebase is a Read tool with low risk. Read-only tools are generally safe to allow by default.
Yes. Add a rate_limit block to the search_codebase rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for search_codebase. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
search_codebase is provided by the Deepcontext MCP server (@wildcard-ai/deepcontext). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Deterministic rules across all 4 Deepcontext tools. Per-identity grants. Full audit log. Live in minutes. Nothing to install.
Free to start. No card required.
4,600+ MCP servers and 31,000+ tools scanned and risk-classified.