Unblocked Context Search — Code

Code-only retrieval across the current repo and other connected repos. Calls context_search_code with a natural-language query to semantically search code — finding implementations, usages, and patterns that Grep can't reach because the code lives in another repo or the query is conceptual rather than literal.

Sources: GitHub, GitHub Enterprise, GitLab, GitLab Self-Managed, Bitbucket, Bitbucket Data Center, Azure DevOps.

How to Invoke

context_search_code is CLI-only in most environments — it does not appear in the MCP tool list even when fully available. Run command -v unblocked once per session and cache the result. Do not conclude the tool is unavailable from the MCP surface alone. See unblocked-tools-guide for full routing rules.

CLI (preferred):

unblocked context-search-code --query "<your query>" [--instruction "<instruction>"]

MCP fallback (only if CLI is confirmed unavailable): fall back to context_research with instruction: "Prefer code and implementation results; deprioritize docs, issues, and messages".

If neither is available: stop and tell the user Unblocked is not configured in this environment (see unblocked-tools-guide for the full message). Do not substitute with other code-search tools.

When This Adds Value Over Grep/Read

Grep and Read are fast for literal matches in the local workspace. Use this tool when:

• The code might live elsewhere — a referenced class, service, or utility may be in a connected repo that isn't in the local workspace
• The query is conceptual — "find all payment processors" or "how is retry logic implemented" won't match a grep pattern
• You want code-only results — context_research returns everything (Slack, Jira, docs, PRs); this returns only code, so results stay focused when that's all you need

If local Grep can find it, use that. This tool's value is reach (cross-repo) and semantics (concept-level matching).

When to Use context_research Instead

Use context_research when you need the full picture alongside the code — PR discussions, team decisions, issue tracker context, Slack threads, architectural docs. This tool returns code only; organizational context requires context_research.

Input

Parameter	Required	Description
query	Yes	What to find — write a complete phrase with concrete identifiers, not bare keywords.
instruction	No	Fine-grained control over which results surface: preferred repos, file types, or code patterns to prioritize or deprioritize.

Writing effective queries — include the most concrete identifiers you have:

• Class or method names (PaymentProcessor.charge(), AuthService)
• File paths or module names (billing-service/src/webhooks, shared/utils/retry)
• Repo names when you know them (payments-service, web-client)
• Conceptual descriptions when you don't (the rate limiting middleware in the API gateway)

Write a complete question or directive, not a keyword fragment:

Instead of	Write
retry logic	Where is retry-with-backoff implemented across connected repos?
auth middleware	Find the authentication middleware that validates JWTs in the API gateway.
PaymentProcessor	How is PaymentProcessor.charge() implemented, and which services call it?
pagination	How do connected repos implement cursor-based pagination in REST APIs?

instruction examples:

• "Prefer interface definitions and entry points over test files"
• "Focus results on the billing-service and payments-service repos"
• "Deprioritize generated or vendor code"

Splitting Queries

Split distinct unknowns into separate context_search_code calls rather than cramming everything into one query. Run them in parallel when the unknowns are independent.

One query, two unknowns (diluted results):

Find the rate limiting middleware and also how authentication tokens are validated.

Two parallel queries (focused results):

Query 1: Where is the rate limiting middleware implemented in the API gateway?

Query 2: How does the authentication service validate JWT tokens?

When to Skip

• Local Grep/Glob/Read finds what you need — no reason to reach for this tool
• You also need PR history, team discussions, issue context, or docs — use context_research instead, which covers all sources in one call
• You already know the exact file and line — direct file reads are faster

Interpreting Results

• Returned code reflects the default branch of each repo, not your local workspace state — verify against local files before acting on the result
• Mine concrete identifiers (file paths, class names, module names, repo names) from results and use them in follow-up queries for precision
• If results are thin, re-query with a more specific identifier you extracted from the first pass
• Cross-reference key claims against the actual source file before driving implementation decisions

Reference

• references/query-patterns.md — query examples organized by use case, with good/bad comparisons and instruction patterns