90 tasks · Claude Opus 4.7 · paired runs

We gave Claude a map
of your codebase.
It stopped getting lost.

Token Savior is an MCP server that indexes a project into a symbol graph — callers, dependents, communities, import cycles, change impact. We ran the same model against the same 90 tasks, first with Read/Grep/Bash, then with Token Savior wired in. Here is what changed.

fewer active tokens

faster wall time

accuracy gained

40 / 48 / 2

wins · ties · losses

the scorecard · what the numbers say

Same model. Same tasks.
One of them kept winning.

Two agents, one codebase, ninety prompts pulled from real engineering work — find callers, audit cycles, estimate blast radius, summarize modules. We counted tokens, wall time and correctness.

agent a · plain

Read, Grep, Bash.

no structural tools

active tokens1.55M

wall time166 min

accuracy66.7 %

score120 / 180

66.7%vs97.8%

accuracy i

40 wins48 ties · TS −47% tokensi2 losses

agent b · token savior

Plus the graph.

+ 59 structural MCP tools (lean profile)

active tokens821k

wall time36 min

accuracy97.8 %

score176 / 180

Accuracy, by category

avg score out of 2 · red = plain · green = token savior

Where the gap opens

context ingested per category · lower is better

−99 %
Import cycle detection
Plain agent ran 35 Bash, 20 Read calls, 113k chars, 132s. Token Savior: two MCP calls, 200 chars, 17s. Same correct answer.
0 → 2
File dependency recovery
Baseline read the whole file and missed three imports. One get_file_dependencies() call got a perfect 2/2.
21
Impossible without
Tasks the plain agent scored 0 on but Token Savior solved — community detection, hotspot audits, semantic duplicates. Amber tiles below.

the graph · what Token Savior sees

A codebase isn't a folder.
It's a graph.

What you see below is the tsbench fixture parsed into its call graph: 206 symbols (functions, classes, methods, constants), 412 call edges connecting them, clustered by module. The plain agent walks this one grep at a time; Token Savior queries it directly. Drag to orbit, hover a point for details.

function class method const 206 symbols 412 call edges drag · scroll · hover

the replay · watch a task resolve

Don't take our word for it.
Watch both agents work.

Every tile is a real task from the benchmark, colored by outcome. Click one and we'll replay both agents' actual traces side by side — real tools, real timings, real answers.

Token Savior wins (11) impossible without TS (21) tie (28) losses (0)

TASK-026 Detect import cycles in the project. Any circular dependencies? If yes, list them.

wall time i

plain—

→

ts—

—

active tokens i

plain—

→

ts—

—

context chars i

plain—

→

ts—

—

score i

plain—

→

ts—

—

agent a · plain

Read · Grep · Bash

agent b · token savior

+ structural MCP

We gave Claude a mapof your codebase.It stopped getting lost.

Same model. Same tasks.One of them kept winning.

Accuracy, by category

Where the gap opens

A codebase isn't a folder.It's a graph.

Don't take our word for it.Watch both agents work.

Stop guessing.Give your agent the map.

We gave Claude a map
of your codebase.
It stopped getting lost.

Same model. Same tasks.
One of them kept winning.

A codebase isn't a folder.
It's a graph.

Don't take our word for it.
Watch both agents work.

Stop guessing.
Give your agent the map.