Pass 3 of 3 · Deep dive of PHAROSPass 1 · OverviewPass 2 · How it works
Code Observatory / Interface
MCP server
Eight read-only tools over streamable HTTP.
Recorded run
PHAROS at 975a38dMIT license
How it works
Assistants connect directly. Every tool is read-only and returns what the REST API returns.
The recorded handshake listed retrieve_context, search_code, find_symbol, list_repositories, get_repo_brief, search_memory, list_skills and get_skill.
Recorded runs through it
- A question reaches verified sourceOne question, asked the way an assistant asks it: a real MCP call, a real dense search, and five files fetched from GitHub and checked against their Git blobs.
- Who calls this, and what does it call?An assistant asks where one function lives and what touches it. A real MCP call finds the definition, then reads the resolved call graph around it.
Source evidence
- The MCP tool
backend/app/modules/mcp/native_server.py:88–126Clamps its limits, then calls the same assemble_context as REST. - The find_symbol tool
backend/app/modules/mcp/native_server.py:170–187Takes a name, an optional repository and a limit. It returns each definition with its graph. - Constant-time token check
backend/app/shared/auth.py:102–110Two static bearer tokens, compared with hmac.compare_digest. Unset tokens fail closed. - What the read token may do
backend/app/shared/auth.py:126–133The read token gets GETs and a short list of read-only POSTs, the MCP endpoint among them.
Claims this supports
Native MCP exposes search, context, symbol, repository and repository-memory tools.
In a recorded run at commit 975a38d, a question sent through native MCP returned five chunks, each fetched from GitHub and matched against its indexed Git blob.
One local run with a question chosen by the author. Not a benchmark and not the deployed service.
