jDocmunch MCP — Claude Code MCP server: section-level doc search
Hands your agent one documentation section instead of a whole file. Useful when the reference set is large and context fills up with text nobody asked for.
claude mcp add jdocmunch-mcp -- uvx jdocmunch-mcpWhat jDocmunch MCP does
jDocmunch parses a documentation set into a tree of sections keyed by heading hierarchy, then records each section's byte offsets in the source file. An agent can ask for a single section, a heading subtree, or a ranked search hit, and gets back just that slice. Section identities survive re-indexing as long as the path, heading text and heading level stay the same.
Setup is a normal MCP install: uv tool install jdocmunch-mcp, then jdocmunch-mcp init. The init step finds your MCP clients, writes their config entries, adds a prompt policy so the agent actually reaches for the tools, and can install hooks and index docs. You need Python 3.10+ and an MCP-compatible client. Hooks are the exception to the zero-install path: a minimal-PATH subshell resolves the executable by name, so they need uv tool install (or pipx/pip).
Index a local folder with jdocmunch-mcp index-local --path ./docs. The JSON reply reports file and section counts; sections outnumber files because the index addresses headings, not files. Installing the server alone won't stop an agent from brute-reading files. One line in CLAUDE.md does that — call jdocmunch_guide and follow it.
What it helps with
- pull one relevant section out of a large doc set without opening the file
- fetch a heading subtree for a specific topic
- index a docs folder and search across sections
- cut context use when answering a configuration question
- keep stable references to sections across re-indexing
Example prompts for the agent
- Search the docs for the authentication setup section and show it to me
- Index the
./docsfolder and tell me how many sections it found - Find the section on node affinity and return only that part
How to install jDocmunch MCP
Platform CLI
claude mcp add jdocmunch-mcp -- uvx jdocmunch-mcpuvx runs the command named after the package; if the README names another one, use uvx --from jdocmunch-mcp <command>
Platform CLI
codex mcp add jdocmunch-mcp -- uvx jdocmunch-mcpuvx runs the command named after the package; if the README names another one, use uvx --from jdocmunch-mcp <command>
Config file
{
"mcpServers": {
"jdocmunch-mcp": {
"command": "uvx",
"args": [
"jdocmunch-mcp"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from jdocmunch-mcp <command>
Config file
{
"servers": {
"jdocmunch-mcp": {
"type": "stdio",
"command": "uvx",
"args": [
"jdocmunch-mcp"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from jdocmunch-mcp <command>
Platform CLI
gemini mcp add jdocmunch-mcp uvx jdocmunch-mcpuvx runs the command named after the package; if the README names another one, use uvx --from jdocmunch-mcp <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about jDocmunch MCP
How do I install jDocmunch MCP in Claude Code?
Run uv tool install jdocmunch-mcp, then jdocmunch-mcp init, which writes config entries for the MCP clients it finds. For a manual Claude setup use claude mcp add jdocmunch-mcp -- uvx jdocmunch-mcp.
What does jDocmunch MCP do?
It indexes documentation by heading hierarchy and serves an agent a single section or subtree, extracted from the original file by byte offsets.
What does the server need to run?
Python 3.10+ and an MCP-compatible client. Hooks require an install via uv tool install, pipx or pip, since they are spawned by a subshell with a minimal PATH.
