Interlinked.

Agents · measured July 2026 · 6 min

Why Claude Code makes so many tool calls.

The short answer: it has no map of your machine. Without an index, an agent can't ask where is this file. It can only ask is it here?, one folder, one pattern, one file at a time. So it lists, greps, reads, and repeats. On a real 4.47-million-file drive we watched that loop run 71 times over 6 minutes 57 seconds to open a single file. Here is the loop, drawn, and why one indexed call ends it.

One file · one session

71

tool calls to open it

6m 57swall clock, spent hunting
~58%of the session's tokens burned
71 → 1calls, once it has an index
→ one indexed call: 16 ms

The count, drawn

One call only rules out one place.

An agent's tools are the ones a person types into a terminal: ls to list, grep to scan, read to open. Each answers a yes-or-no about one spot on the disk. To locate one file among millions with only local yes-or-no probes, you need a lot of probes: about eighteen times through the same four-step round. And every round re-reads the whole transcript so far, so the calls get more expensive as they pile up. That is why the count, and the token bill, climbs the way it does.

WITHOUT AN INDEX · ONE SESSION TO OPEN ONE FILElistgrepreaddecide= one tool call eachevery call re-reads the growing transcript: later calls cost moreone round → still not sure → go againfound it · call 71~58%SESSION TOKENS SPENT FINDING THE FILE, NOT CHANGING IT71 CALLS · 6M 57SVSWITH AN INDEX · ONE CALLone lookup: no walk, no retries, no transcript re-reads~0%1 CALL · 16 MS · ~0 TOKENS

Same machine, same drive, same file. Seventy-one blind probes on top; one indexed call on the bottom.

One real task, measured

What 71 calls actually cost.

71 → 1
tool calls the agent made to find one file
6m 57s → 16ms
wall clock: the hunt, then one indexed call
~58% → ~0%
of session tokens spent finding, not building
~7,200,000×

The distance between hunting for a file and knowing where it is, on the same drive. A narrower, cleaner test (one file by name, nothing else) put Claude Code at 193.5 seconds against the same indexed lookup at 85 microseconds: 2,276,000×.

This was Claude Code running Anthropic's strongest agent. The $200-a-month plan buys a better model, not a better floor: the tools underneath stay the same ls, grep and read, so the smartest agent still spends most of a session finding things instead of building them. Measured on a Ryzen 9 9950X3D, 64 GB DDR5, NVMe, Windows 11, 4.47M files.

Same machine · same drive · same jobs

The two famous anchors, lined up on one drive.

Claude Code is the agent making the calls; ripgrep is the fastest grep there is, and it is what the agent shells out to for content. ripgrep is not slow: it is doing honest work fast. It just re-walks the disk on every query, because there is nothing warm to ask.

The searcherThe jobTimevs Interlinked
Interlinkedone indexed call, filename or content85 µs / 16 msthe baseline
Claude Code (Opus, single lookup)the 71-call hunt for one file193.5 s2,276,000×
ripgrepgrep one string across the drive93.8 s58,625× (5-query set)
VS Code (Ctrl+P)open a file by name35 s412,000×
Windows Searchfind a file by name (median of 20)67 s~480,000×

Methodology: Ryzen 9 9950X3D · 64 GB DDR5 · NVMe · Windows 11 · 4,470,000 files. Filename lookups are 85 µs single (2 µs best case) and 139 µs median of 20 queries. Content search averages ~9 ms across a five-query set; ripgrep averaged 58,625× slower on that set, and on one representative query took 93.8 s to Interlinked's 16 ms. VS Code Ctrl+P (412,000×) and Windows Search (~480,000×) are the file-open and median-of-20 times on the same drive.

It's not the model

The tools were built for one open project, not a whole machine.

VS Code indexes the folder you opened. Cursor indexes the repo you're in. Both are fast inside that one boundary. But an agent's questions don't respect folder boundaries: the answer might be in a sibling repo that ships alongside this one, or in the 90% of the disk that was never in git: a config, a document, a log. Nothing keeps a warm, machine-wide view of all of it.

So the agent falls back to the floor: ripgrep and ls, one blind probe at a time, because nothing across the machine is kept ready between them. Give Opus a map and it makes one call; give it a flashlight and it makes seventy-one. The count is a property of the tools, not the model.

The fix

One call. The other seventy are gone.

Interlinked keeps a live index of every file on the machine (names and contents) and hands it to any agent over MCP. One signed install auto-configures 19 AI clients, Claude Code among them. Once the indexed-search tool is registered, the agent stops reaching for grep: it asks where is this once and gets the exact path back. No walk, no round, no retry. The index sits warm in the background at about 44 MB and stays current as you save (about a millisecond, under 30 ms worst case), so the answer is never stale.

Same agent · same model · same query

Without index

Calls
71
Time
6m 57s
Tokens
~58%

With Interlinked

Calls
1
Time
16 ms
Tokens
~0

Seventy-one blind probes replaced by a single index lookup. Same model, same machine, a different floor.

1.7 s
day one, a fresh machine's 12 of 12 repos were webbed: 26,958 relationship pairs from its own history, no model involved
~44 MB
idle footprint in Task Manager, less than a browser tab; the index on disk stays under 1% of the drive
19
AI clients configured by one install: Claude Code, Cursor, Codex, Copilot, Windsurf, Zed and the rest

THE MODEL DIDN'T GET SMARTER. THE FLOOR GOT FASTER. THAT'S THE ONLY THING THAT CHANGED.

Do this today

Turn 71 calls into one.

1

Install once. One signed installer auto-configures 19 AI clients: Claude Code, Cursor, Copilot, Codex, Windsurf, Zed and the rest. No JSON editing.

2

Let it index. Filename search works immediately; content search fills in behind it. At rest the whole engine sits around 44 MB and updates as you save.

3

Scope when you can. Whole-machine when you don't know where something lives; scoped to a folder when you do. Scoped queries return in the microsecond range.

4

Watch the count drop. The agent trades seventy-one blind probes for one call that returns the exact path, and stops spending its token budget on directory listings.

Everything local is free forever, on all your devices, no card. Hosting starts at $5.99/mo. Download for Windows.

Measured July 2026 · Ryzen 9 9950X3D · 4.47M files · Windows 11← All research