41 lines
1.8 KiB
Markdown
41 lines
1.8 KiB
Markdown
|
# Product Context: PDF Reader MCP Server
|
||
|
|
|
||
|
|
## 1. Problem Solved
|
||
|
|
|
||
|
|
AI agents often need to access information contained within PDF documents as
|
||
|
|
part of user tasks (e.g., summarizing reports, extracting data from invoices,
|
||
|
|
referencing documentation). Directly providing PDF file content to the agent is
|
||
|
|
inefficient (large token count) and often impossible due to binary format.
|
||
|
|
Executing external CLI tools for each PDF interaction can be slow, insecure, and
|
||
|
|
lack structured output.
|
||
|
|
|
||
|
|
This MCP server provides a secure, efficient, and structured way for agents to
|
||
|
|
interact with PDF files within the user's project context.
|
||
|
|
|
||
|
|
## 2. How It Should Work
|
||
|
|
|
||
|
|
- The server runs as a background process, managed by the agent's host
|
||
|
|
environment.
|
||
|
|
- The host environment ensures the server is launched with its working directory
|
||
|
|
set to the user's current project root.
|
||
|
|
- The agent uses MCP calls to invoke specific PDF reading tools provided by the
|
||
|
|
server.
|
||
|
|
- The agent provides the relative path to the target PDF file within the project
|
||
|
|
root.
|
||
|
|
- The server uses the `pdf-parse` library to process the PDF.
|
||
|
|
- The server returns structured data (text, metadata, page count) back to the
|
||
|
|
agent via MCP.
|
||
|
|
- All file access is strictly limited to the project root directory.
|
||
|
|
|
||
|
|
## 3. User Experience Goals
|
||
|
|
|
||
|
|
- **Seamless Integration:** The agent should be able to use the PDF tools
|
||
|
|
naturally as part of its workflow without complex setup for the end-user.
|
||
|
|
- **Reliability:** Tools should reliably parse standard PDF files and return
|
||
|
|
accurate information or clear error messages.
|
||
|
|
- **Security:** Users should trust that the server only accesses files within
|
||
|
|
the intended project scope.
|
||
|
|
- **Efficiency:** Reading PDF data should be reasonably fast and avoid excessive
|
||
|
|
token usage compared to sending raw file content (which isn't feasible
|
||
|
|
anyway).
|