navigation-mcp

name	navigation-mcp
description	Trigger: MCP navigation, code discovery, symbol lookup, trace_flow, trace_callers, endpoints. Use navigation-agent MCP before reading workspace code.
license	Apache-2.0
compatibility	opencode
metadata	{"author":"j0k3r-dev-rgl","version":"1.7.0"}

Activation Contract

Use this skill for workspace-only code discovery: locating files, symbols, routes, endpoints, text matches, caller impact, or downstream flow before editing. Do not use it for web search, external repositories, parent directories, arbitrary filesystem access, or reading full file contents.

Hard Rules

1. If discovery is needed, use navigation first; do not open multiple files cold.
2. If the exact file is already known and no discovery is needed, read is allowed.
3. Canonical tool names are code.inspect_tree, code.find_symbol, code.list_endpoints, code.search_text, code.trace_flow, and code.trace_callers.
4. Clients may expose aliases such as navigation-agent_code_find_symbol; use the available alias but preserve the canonical code.* semantics.
5. Supported language filters: typescript, javascript, go, java, php, python, rust, csharp.
6. Supported framework filters: react-router, spring.
7. Use snake_case parameters (max_depth, include_hidden, file_pattern, recursive).
8. Treat navigation output as a scope reducer, not file content. Read only returned files that matter.
9. Use glob, bash grep, or bash find only when navigation is unavailable, unsupported for the task, or returned no useful result after narrowing.

Decision Gates

Need	Use first	Then
Understand an unknown module	code.inspect_tree	Read selected paths only
Find a class/function/method/type	code.find_symbol	Trace or read returned path
Audit REST/GraphQL/routes	code.list_endpoints	Trace or read entrypoints
Find imports/decorators/text	code.search_text	Prioritize topFiles, then read
Know what a symbol calls	code.find_symbol	code.trace_flow with items[].path
Know who calls a symbol	code.find_symbol	code.trace_callers with recursive: true for shared APIs

Execution Steps

1. Identify the narrowest known scope: path, language, and framework when available.
2. Start with the most specific navigation tool from the decision table.
3. Chain tool outputs: pass find_symbol.items[].path as path to trace_flow or trace_callers.
4. Remember direction: trace_flow is downstream behavior; trace_callers is upstream impact.
5. If find_symbol misses constants, config keys, decorators, imports, or generated names, use code.search_text scoped by path, include, and language.
6. If trace output is too broad, narrow path, language, framework, or symbol; for trace_callers, lower max_depth.
7. If endpoint discovery returns zero, retry with a narrower path and the most specific framework or kind before concluding there is no public surface.
8. If results are empty or truncated: true, narrow before reading files or increasing limit.
9. Read only the small set of files needed to answer or edit safely.

Output Contract

When reporting back, mention the navigation path used, important resolved files, any fallback, and limitations such as unsupported framework detection or zero endpoint results.

Anti-Patterns

• Reading broad directories before navigation.
• Using text search when a symbol trace would answer better.
• Calling trace_flow to find callers or trace_callers to follow behavior.
• Assuming this MCP can inspect the web or external repos.
• Treating a zero-result navigation response as proof without a scoped fallback.

References

• README.md — public contract, compatibility matrix, and agent usage guide.