My AI Couldn’t See My Information — I Constructed a Zero-Dependency MCP Server

. The features had grown too lengthy and the variable names made no sense anymore. Each time I wished suggestions on a file, I ended, opened the chat, copied the entire thing in, and waited. Then went again to the editor, utilized the change, opened the following file, and did it once more.

Sooner or later I counted. Six information. Eleven pastes. Twenty minutes of switching earlier than I wrote a single new line.

The apparent repair was to offer the AI instrument direct entry to my mission folder. That’s after I bumped into MCP — the Mannequin Context Protocol — which is strictly constructed for this. A server runs domestically, exposes instruments, and the AI shopper calls these instruments straight as a substitute of ready for me to stick issues.

So I checked out current implementations. Most required FastAPI, uvicorn, LangChain, or the official MCP SDK. Earlier than writing a single line of enterprise logic I had 5 packages in my necessities file and a server I wasn’t assured would run on Home windows with no combat.

I stepped again and browse the precise MCP spec [1]. The protocol is JSON-RPC 2.0 [2] over a transport layer. One JSON object per line. Shopper sends, server responds. The spec defines precisely two transports: stdio for native single-client connections, and HTTP with Server-Despatched Occasions for concurrent shoppers.

That’s the entire protocol.

I requested a unique query: what does this really need that Python’s customary library doesn’t already present? sys.stdin, sys.stdout, http.server, threading, queue, pathlib, json. That’s it. Not a single pip set up.

This text is that implementation — each transports, a manufacturing safety mannequin, 50 assessments, and the numbers from working it.

TL;DR

Most MCP implementations really feel heavier than they need to. The spec solely defines two transports, stdio and HTTP/SSE, however in follow they’re normally wrapped in frameworks and further dependencies.

I constructed each transports from scratch utilizing solely the Python customary library.

It runs as a single file with one runtime flag. No installs, no setup.

For native work, it makes use of stdio with a single shopper. Whenever you want concurrency, it switches to HTTP/SSE and handles a number of shoppers with out altering anything.

Underneath the hood, every part stays constant. Identical dispatcher, identical instruments, identical safety mannequin.

As a result of it touches the filesystem, I added strict path checks early on. Frequent escape patterns like ../../, symlink tips, and Home windows UNC paths are blocked.

5 concurrent shoppers. Underneath 50ms whole wall time. Verified on Home windows 11, Python 3.12.6, CPU solely.

Full code: https://github.com/Emmimal/local-mcp-server/

The Mistake That Formed the Entire Design

Earlier than the structure, I need to inform you concerning the factor that just about made me hand over on the entire thing.

Early in improvement I used to be testing the search instrument. I pointed the server at C:UsersAdmin and ran it in search of Python information. The server began. The demo began working. Then it simply stored working.

Thirty seconds. A minute. 5 minutes. I believed there was an infinite loop someplace. I went again via the code 3 times. All the things seemed appropriate. I killed the method and restarted. Identical outcome.

Ten minutes in I lastly understood what was taking place. The search instrument was utilizing rglob() by default. I had pointed it at my total consumer listing and it was scanning every part — digital environments, AppData, each cached file on the machine. Tens of hundreds of information, one by one.

I killed the method and altered one line:

# Earlier than — recursive by default, scans every part
for match in goal.rglob(sample):

# After — shallow by default, opt-in for recursion
for match in goal.glob(sample):

And made recursive=False the default parameter. The shopper has to cross recursive=True explicitly. The server won’t ever scan recursively by itself.

That single change is why search completes in below 30ms on an actual mission folder right this moment as a substitute of working without end. And it turned the rule I utilized in all places: no habits that destroys efficiency ought to ever be the default.

What MCP Truly Is

The Mannequin Context Protocol [1] is a standardised method for AI shoppers to name instruments on exterior servers. It makes use of JSON-RPC 2.0 [2] as its message format.

In follow, this implies AI shoppers like Claude or ChatGPT can straight entry and cause over native information as a substitute of counting on copy-paste.

The handshake has three phases. First the shopper initializes, then it asks what instruments can be found, then it begins calling them:

All the things after that’s the transport carrying messages forwards and backwards.

The spec defines two transports. stdio runs over customary enter and output — one JSON object per line, flushed instantly. HTTP/SSE runs requests over HTTP POST, with responses streamed again over a persistent Server-Despatched Occasions connection [3].

Most implementations decide one. This one implements each, with the identical dispatcher and the identical 4 instruments sitting behind every.

Here’s what the demo reveals at startup — each transports register the identical instruments:

[2] Accessible instruments
  [list_directory     ] Listing information and directories. Returns title, sort, dimension...
  [read_file          ] Learn a file's contents. Max 1 MB. Binary information returned...
  [search_files       ] Search information by glob sample. Use recursive=true for...
  [get_file_info      ] Get metadata for a file or listing: dimension, sort, ext...

Structure: 4 Layers

The system has 4 layers.

Safety layer — validates each path earlier than any filesystem operation. It runs earlier than anything, on each single name.

Instruments layer — 4 instruments for the precise file system work: list_directory, read_file, search_files, get_file_info.

Dispatcher — a stateless JSON-RPC router. Parses the tactic, calls the precise handler, returns the response. It has no concept which transport is working and it doesn’t must.

Transport layer — two implementations. StdioTransport for native AI shoppers. HTTPSSETransport for concurrent connections. The dispatcher has no concept which one is working.

The entry level selects the transport at startup:

dispatcher = MCPDispatcher(root)

if args.http:
    HTTPSSETransport(dispatcher, host=args.host, port=args.port).run()
else:
    StdioTransport(dispatcher).run()

One flag. That’s it.

The Safety Mannequin

The very first thing I had to consider when constructing a server that reads native information was what stops a shopper from studying information it shouldn’t. The apparent assault is path traversal — as a substitute of sending README.md, a shopper sends ../../and so forth/passwd and a server that doesn’t examine follows it straight out of the sandbox.

The repair was to resolve each paths absolutely earlier than evaluating them. The important thing line:

goal.resolve().relative_to(base.resolve())

Path.resolve() expands all symlinks and collapses all .. segments. relative_to() raises ValueError if the outcome lands exterior the bottom. [6] No string parsing, no counting .. manually. The OS resolves the trail; Python checks the outcome.

MCP_ROOT units the sandbox root by way of setting variable. I set it to my mission folder particularly, not my residence listing. Each instrument runs this examine earlier than touching the filesystem. If it fails, the error goes again to the shopper instantly.

The safety assessments confirm this on each construct:

The 4 Instruments

list_directory

Lists every part in a listing — title, sort, dimension, modified timestamp, relative path. Directories earlier than information, hidden entries excluded by default.

Pointing it on the mission folder:

[3] list_directory
  8 entries:

  [F] concurrent_demo.py                  4,711B
  [F] demo.py                            10,451B
  [F] http_client.py                      5,140B
  [F] local_desktop_config.json             228B
  [F] README.md                           7,542B
  [F] server.py                          29,222B
  [F] test_server.py                     17,500B

Eight entries, sizes, all contained in the sandbox. The type order places directories first as a result of the kind key makes use of p.is_file() — False < True in Python, so directories naturally float up.

One factor that bit me on Home windows: a file can seem in a listing itemizing whereas being locked by one other course of. merchandise.stat() raises PermissionError on that entry. The instrument wraps every stat name in its personal attempt/besides and skips locked entries silently as a substitute of crashing your entire itemizing.

read_file

Reads file contents with a tough 1 MB cap. Textual content information returned as plain UTF-8. Binary information returned as base64.

read_file
  concurrent_demo.py:

  #!/usr/bin/env python3
  """
  concurrent_demo.py
  ============================
  Proves the HTTP/SSE transport handles a number of concurrent shoppers.

  Spins up 5 shoppers concurrently, every working
  ... (4509 extra chars)

I added the binary fallback after pointing the server at an actual mission folder for the primary time. Python mission folders include .pyc information, compiled extensions, SQLite databases. The primary model refused all of them with UnicodeDecodeError. The repair: if read_text() fails on decode, fall again to read_bytes() and return base64. The shopper will get a structured response with a binary: true flag as a substitute of a crash.

The 1 MB cap exists as a result of one early take a look at by accident learn a 200 MB SQLite database and froze the method for thirty seconds. MAX_FILE_BYTES is a continuing on the high of server.py — change it in case your workflow wants bigger information.

search_files

After the rglob() incident, this instrument works like this:

[6] search_files — *.py (shallow)
  Discovered 5 file(s):

  -> concurrent_demo.py    4,711B
  -> demo.py              10,451B
  -> http_client.py        5,140B
  -> server.py            29,222B
  -> test_server.py        17,500B

5 information, below 30ms. The identical name on C:UsersAdmin with recursive=True would nonetheless scan every part — however now that may be a deliberate selection the shopper has to make, not one thing the server does routinely.

The truncated flag tells the shopper when outcomes had been minimize off at max_results. The primary model silently dropped outcomes with no sign — I added truncated after realising the shopper had no solution to understand it wasn’t getting every part.

get_file_info

Returns metadata with out studying file contents — helpful when the shopper must examine permissions earlier than deciding whether or not to learn.

[4] get_file_info
  title         local-mcp-server
  path         .
  sort         listing
  dimension         4096
  modified     1780246573
  created      1780227648
  extension    None
  readable     True
  writable     True

os.entry() checks actual permissions, not simply existence. On Home windows a file could be seen in a list whereas being locked. Understanding it’s unreadable earlier than making an attempt to learn it saves a spherical journey.

The Dispatcher

I didn’t need to reinvent the wheel or rewrite my core logic simply to deal with totally different community setups, so I constructed a central dispatcher to deal with every part as a substitute. It features as a fundamental, stateless engine. A uncooked JSON string is available in, the dispatcher parses it to see precisely what the shopper wants, after which it drops a response again.

I explicitly stored all community and file I/O out of this element. It doesn’t know something about stdin, stdout, or HTTP. All of that messy communication is left fully to the transport layers. The transports do the heavy lifting with the precise sockets or streams and easily cross the clear knowledge alongside to the dispatch() operate.

To maintain the system lean, the engine solely listens for 4 spec strategies: initialize, instruments/checklist, instruments/name, and ping. If anything hits the dispatcher, it shuts the request down instantly with a regular JSON-RPC error.

The one exception is dealing with notifications. When a message comes via with out an id discipline, the MCP specification dictates that no response is required. The dispatcher processes the occasion internally and simply returns None. As a result of the core engine is totally impartial of how knowledge travels, transferring from native stdio to an HTTP server requires zero inner code adjustments. The transport layer adjustments on the skin, however the primary dispatcher stays precisely the identical.

Transport 1: stdio

For the native setup, the stdio transport is only a uncooked for line in self._stdin loop. I fully skipped async, threads, and occasion loops to maintain it so simple as attainable.

The Home windows repair truly took me longer than writing the transport itself. By default, Python opens stdin and stdout in textual content mode on Home windows, which routinely adjustments each n to rn everytime you write knowledge. That little change fully corrupts the JSON stream. The second a shopper reads }rn{, it hits a parse error on the very subsequent message, breaking your entire connection.

if platform.system() == "Home windows":
    import msvcrt
    msvcrt.setmode(sys.stdin.fileno(),  os.O_BINARY)
    msvcrt.setmode(sys.stdout.fileno(), os.O_BINARY)

Setting O_BINARY disables the interpretation. [8] With out this the server works on macOS and Linux and silently breaks on Home windows.

write_through=True on the stdout wrapper ensures each write flushes instantly. The AI shopper is obstructing synchronously ready for the response — any buffering stalls the interplay.

Right here is the total stdio demo output from my machine:

============================================================
  local-mcp-server demo  [stdio transport]
  Root: C:UsersAdminPycharmProjectspythonProjectlocal-mcp-server
============================================================

[1] Initialize
  Server  : local-mcp-server v1.0.0
  Protocol: 2024-11-05

[2] Accessible instruments
  [list_directory     ] Listing information and directories...
  [read_file          ] Learn a file's contents. Max 1 MB...
  [search_files       ] Search information by glob sample...
  [get_file_info      ] Get metadata for a file or listing...

[3] list_directory     8 entries
[4] get_file_info      readable: True  writable: True
[5] read_file          first small file learn efficiently
[6] search_files       Discovered 5 .py information

============================================================
  All checks handed. Prepared to attach Native Desktop.
============================================================

Transport 2: HTTP/SSE

Every shopper opens a GET /sse connection (constructed on Python’s http.server [4]) that stays open for your entire length of the session, permitting the server to push responses down that pipeline as server-sent occasions. Every connection receives a singular client_id [9] on join. When a shopper wants to speak again or ship a request, it fires off a separate POST /message.

The move per shopper appears to be like like this:

To deal with concurrency cleanly, every shopper will get its personal impartial message queue. [7] The POST handler dispatches the decision, drops the outcome straight onto that shopper’s queue, and instantly returns a 202 standing. It doesn’t watch for the SSE supply to complete. The shopper simply picks up the response from its personal open stream. That’s what makes the concurrency work.

I arrange 16 daemon employee threads to handle incoming requests. Since every energetic SSE connection holds onto one thread, having 5 energetic SSE shoppers leaves 11 threads fully free to deal with incoming POST requests at any second. There isn’t a async/await syntax and no occasion loop—simply customary library threading. [5]

The Concurrent Demo

That is the output that solutions whether or not the HTTP/SSE transport truly works:

============================================================
  Concurrent Shopper Demo — 5 shoppers, 5 simultaneous calls
============================================================

  Launching 5 concurrent shoppers...

  Shopper     Instrument                 Consequence         Time
  ---------- -------------------- ---------- --------
  1          list_directory       OK           ~0.034s
  2          get_file_info        OK           ~0.021s
  3          list_directory       OK           ~0.038s
  4          search_files         OK           ~0.023s
  5          search_files         OK           ~0.021s

Complete wall time: ~0.04s for five concurrent shoppers
Consequence: ALL PASSED
============================================================

5 shoppers. 5 totally different instrument calls. Underneath 50ms whole wall time throughout all runs. None blocked one another. Measured on Home windows 11, Python 3.12.6, CPU solely.

What Broke Throughout Improvement

The ten-minute cling I already described. Three different issues broke earlier than the server was steady.

The Home windows rn downside. The primary time I related an precise AI shopper it acquired a parse error on the second message. All the things seemed nice in testing. The difficulty was the stdout translation — n turning into rn on Home windows. I spent an hour wanting on the dispatcher earlier than I discovered it. Two strains fastened it.

The binary file crash. First model of read_file known as read_text() on every part. First actual mission folder it hit a .pyc file and raised UnicodeDecodeError. Added the base64 fallback after that.

The 200 MB database freeze. Earlier than the 1 MB cap, a take a look at by accident learn a SQLite database. The method froze for thirty seconds. The cap went in instantly after.

Every of those solely appeared when the server ran towards an actual machine, not a take a look at listing.

The Take a look at Suite

50 assessments throughout seven lessons. Safety runs first.

Run the take a look at suite with pytest in verbose mode. To skip integration assessments, cross the not integration marker flag.

Connecting to a Native AI Shopper

macOS: ~/Library/Utility Assist/Claude/local_desktop_config.json

Home windows: %APPDATApercentClaudelocal_desktop_config.json

{
  "mcpServers": {
    "local-desktop": {
      "command": "python",
      "args": ["C:/absolute/path/to/local-mcp-server/server.py"],
      "env": {
        "MCP_ROOT": "C:/absolute/path/to/your/workspace"
      }
    }
  }
}

For HTTP/SSE:

# Terminal 1 — begin the server
python server.py --http --port 8765

# Terminal 2 — run the instance shopper
python examples/http_client.py

Sincere Design Selections

A pool of 16 employee threads is lots for native improvement, however I didn’t design this to scale right into a shared server dealing with tons of of simultaneous connections. When you want that sort of scale, you need to in all probability swap this out for asyncio and a devoted async framework. For native AI tooling working a handful of shoppers by yourself machine, 16 threads is greater than sufficient.

The safety mannequin trusts the sandbox boundary itself, fully ignoring file varieties. I didn’t write an allowlist of secure extensions or a blocklist of harmful ones. If a path resolves inside MCP_ROOT, it’s readable. One rule is tougher to get round than ten.

I additionally deliberately unnoticed token counting. This server merely returns uncooked file contents. Managing your token funds belongs within the execution layer between the server and the mannequin. Including a counter right here would pressure a tokenizer dependency—breaking the zero-dependency purpose—or pressure an approximation with its personal messy edge circumstances.

Lastly, search is shallow by default. A ten-minute cling throughout testing made this resolution for me. Any habits that silently destroys efficiency like that ought to by no means be the default choice.

What This Truly Teaches

I anticipated constructing an MCP server to be sophisticated. The tutorials made it look sophisticated. Each implementation I discovered had FastAPI, uvicorn, and three different packages earlier than a single instrument was registered. So I assumed that complexity was vital.

It wasn’t. After I lastly learn the precise spec, the protocol was a loop. Learn a line. Parse JSON. Name a operate. Write a line. That’s it. The frameworks weren’t fixing MCP issues — they had been fixing HTTP issues that MCP over stdio doesn’t have.

The usual library was sufficient as a result of the issue was small. I didn’t want a framework. I wanted http.server for TCP connections, threading for parallel requests, queue to decouple SSE from POST dealing with, and pathlib for path decision. One module per downside. Nothing left over.

The factor that shocked me most was how a lot the defaults mattered. Each actual failure on this codebase — the ten-minute cling, the 200 MB freeze, the Home windows JSON corruption — got here from a default that labored nice in testing and broke on an actual machine. rglob() was nice on a small take a look at folder. Textual content mode stdout was nice on Linux. The default that feels handy in improvement is commonly the one which silently destroys issues in manufacturing.

Full code: https://github.com/Emmimal/local-mcp-server/

References

[1] Mannequin Context Protocol. (n.d.). Mannequin Context Protocol Specification. https://modelcontextprotocol.io

[2] JSON-RPC Working Group. (2010). JSON-RPC 2.0 Specification. https://www.jsonrpc.org/specification

[3] WHATWG. (n.d.). Server-sent occasions. HTML Dwelling Customary. https://html.spec.whatwg.org/multipage/server-sent-events.html

[4] Python Software program Basis. http.server — HTTP servers. Python 3 Documentation. https://docs.python.org/3/library/http.server.html

[5] Python Software program Basis. threading — Thread-based parallelism. Python 3 Documentation. https://docs.python.org/3/library/threading.html

[6] Python Software program Basis. pathlib — Object-oriented filesystem paths. Python 3 Documentation. https://docs.python.org/3/library/pathlib.html

[7] Python Software program Basis. queue — A synchronized queue class. Python 3 Documentation. https://docs.python.org/3/library/queue.html

[8] Python Software program Basis. msvcrt — Helpful routines from the MS VC++ runtime. Python 3 Documentation. https://docs.python.org/3/library/msvcrt.html

[9] Python Software program Basis. (n.d.). uuid — UUID objects in line with RFC 4122. Python 3 Documentation. https://docs.python.org/3/library/uuid.html

[10] Python Software program Basis. subprocess — Subprocess administration. Python 3 Documentation. https://docs.python.org/3/library/subprocess.html

Disclosure

All code on this article was written by me and is authentic work, developed and examined on Python 3.12.6, Home windows 11, CPU solely. No GPU was used at any stage. All benchmark numbers — response occasions, concurrent shopper outcomes, take a look at counts — are from precise runs on my native machine and are absolutely reproducible by cloning the repository and working demo.py and concurrent_demo.py as described above. The whole implementation makes use of solely the Python customary library. No third-party packages are required or used at any level. All structure choices, implementation decisions, design tradeoffs, debugging experiences, and the failures described in “What Broke Throughout Improvement” are my very own. I’ve no monetary relationship with any instrument, library, framework, or firm talked about on this article. The MCP protocol is an open specification printed by Anthropic [1]; this implementation is impartial and isn’t affiliated with or endorsed by Anthropic.

When you construct manufacturing AI methods and need to go deeper — tutorials, studying tracks, and hands-on initiatives at EmiTechLogic, my AI and Python studying platform.