Skip to main content
Glama

Context Mode

A free, local plugin for 11 coding agents. It keeps tool output out of your agent's context window and keeps all usage data on your machine. The plugin is free forever.

npm downloads GitHub stars Discord License: ELv2

34K GitHub stars · 866K npm downloads (read 2026-11-19, stats.json) · on the Hacker News front page and GitHub Trending: charts and sources

The Context Mode plugin is free and local, with its source open on GitHub. Context Mode Gateway is a separate hosted service with its own account and subscription. It takes the same idea from tool output to the whole conversation: Gateway Recall keeps long sessions searchable, and Gateway Memory shares what your agents learn across sessions and tools. See how the Gateway works

Contents: Why · Adoption · Install · Try it · Use cases · Benchmarks · How it works · Tools · Security and privacy · Context Mode Gateway · License · Community

Why

A coding agent reads files, runs commands and fetches pages, and every result goes into its context window. The Context Mode plugin runs that work in a local sandbox, keeps the full output in a SQLite index on your machine, and gives the agent the part it asks for.

  • The plugin is free forever. Use it at home and at work, free of charge, under a source-available license.

  • Runs on your machine. The hooks, the sandbox and search run as local processes next to your agent.

  • The plugin keeps your data on your machine. Every index lives in ~/.context-mode: one context.db per project.

  • Zero telemetry. The plugin keeps all usage data on your machine. Security and privacy lists each network call it makes.

  • Works offline after install. Sandbox runs, file indexing and search use local SQLite and your own runtimes.

  • Code Mode. The agent writes a short script instead of reading the data, and only what the script prints reaches the context: see How it works, step 5.

  • Small replies. On the 36 bench rows with 10 KB or more of raw output, the plugin's median reply is 99.4% smaller than the raw output.

Related MCP server: fittok

Adoption: stars and npm downloads

Open the interactive chart and point at a bar to read its value. Tabs there show days, weeks, months and years. Stars come from the GitHub GraphQL API (starredAt); npm downloads come from api.npmjs.org/downloads/range. The total up to 2026-10-29 is 683K npm downloads (stats.json keeps a daily copy). A daily job redraws both charts.

  • Hacker News. "MCP server that reduces Claude Code context consumption by 98%" reached the front page with 570+ points and 107 comments on 2026-02-28 (thread).

  • GitHub Trending. A bot that watches GitHub Trending logged the repo there on 14 days from 2026-04-24 to 2026-10-17 (uibm/trending-watcher).

Starred by engineers at these companies: each name is the company field of 2 or more stargazer profiles, read with the GitHub GraphQL API on 2026-11-01 (data).

Install

Pick your agent: Claude Code · Codex · Gemini CLI · Qwen Code · Pi · OMP · OpenCode · Copilot CLI · Kiro · VS Code · ZCode

Open its block and copy its own install command. Each block also holds the update, check and remove steps, and the Install guide walks through them.

The plugin needs Node 24.15.0 or later, from the engines field in package.json. On Windows, add Git for Windows: the shell tools run in Git Bash, and doctor checks for it. contextmode installs as the npm package and command context-mode. Claude Code, Codex, Gemini CLI and Qwen Code install from GitHub, and each of their commands names the release branch.

claude plugin marketplace add mksglu/context-mode
claude plugin install context-mode@context-mode
# update
claude plugin update context-mode@context-mode
# check
claude plugin list
# remove
claude plugin marketplace remove context-mode

Hooks ✓ (package) · MCP ✓ · Skills ✓

codex plugin marketplace add mksglu/context-mode
codex plugin add context-mode@context-mode
# one-time step: trust the plugin hooks
npm i -g context-mode
context-mode setup --agents codex
# update
codex plugin marketplace upgrade context-mode
# check
codex mcp list
# remove
codex plugin remove context-mode@context-mode

Hooks ✓ (hook file) · MCP ✓ · Skills ✓

gemini extensions install https://github.com/mksglu/context-mode
# update
gemini extensions update context-mode
# check
GEMINI_CLI_TRUST_WORKSPACE=true gemini mcp list
# remove
gemini extensions uninstall context-mode

Hooks ✓ (package) · MCP ✓ · Skills ✓

qwen extensions install https://github.com/mksglu/context-mode:context-mode
# update
qwen extensions update context-mode
# check
qwen extensions list
# remove
qwen extensions uninstall context-mode

Hooks ✓ (package) · MCP ✓ · Skills ✓

pi install npm:context-mode
# update
pi update --extensions
# check
pi list
# remove
pi remove npm:context-mode

Hooks ✓ (the plugin's own event handlers) · MCP ✓ (ctx tools built into the plugin) · Skills ✓

omp plugin install context-mode
# update
omp plugin upgrade context-mode
# check
omp plugin list
# remove
omp plugin uninstall context-mode

Hooks ✓ (the plugin's own event handlers) · MCP ✓ (ctx tools built into the plugin) · Skills ✓

opencode plugin add context-mode
# update
opencode plugin update
# check
opencode mcp list --pure          # OpenCode 1.19
opencode mcp list --standalone    # OpenCode 2
# remove
opencode plugin remove context-mode

Hooks ✓ (the plugin's own event handlers) · MCP ✓ · Skills ✓

npm i -g context-mode
context-mode setup --agents copilot-cli
# update
context-mode upgrade
# check
context-mode doctor --agents copilot-cli
copilot mcp list
# remove
context-mode setup --remove --agents copilot-cli

Hooks ✓ (hook file) · MCP ✓ · Skills ✓

npm i -g context-mode
context-mode setup --agents kiro
# update
context-mode upgrade
# check
context-mode doctor --agents kiro
# remove
context-mode setup --remove --agents kiro

Hooks ✓ (hook file) · MCP ✓ · Skills ✓

npm i -g context-mode
context-mode setup --agents vscode
# update
context-mode upgrade
# check
context-mode doctor --agents vscode
# remove
context-mode setup --remove --agents vscode

Hooks ✓ (hook file) · MCP ✓ · Skills ✓

zcode plugins marketplace add mksglu/context-mode
zcode plugins install context-mode@context-mode
# update
zcode plugins update context-mode@context-mode
# check
zcode plugins list
# remove
zcode plugins uninstall context-mode@context-mode

Hooks ✓ (hook file) · MCP ✓ · Skills ✓

The npm route works for every agent:

npm i -g context-mode
context-mode setup

Setup is also the repair. It shows a plan, saves your earlier config, then applies the plan; when a step ends early, it restores the saved config. To uninstall:

context-mode setup --remove
npm rm -g context-mode

Restart the agent, then run /ctx-doctor in it. Each check prints PASS or a fix command, and the Doctor guide explains each check.

Every agent gets hooks, the MCP server and the plugin skills; the last line of each block above says how. Hooks reach each agent in one of three ways: in the plugin package (Claude Code, Gemini CLI, Qwen Code), in a hook file (Codex, Copilot CLI, Kiro, VS Code, ZCode), or in a plugin inside the agent (OpenCode, Pi, OMP). The Agents guide shows what each agent runs. We ran the real binary of each agent with the plugin hook: Claude Code, Codex, Gemini CLI, Qwen Code, Pi, OpenCode, Copilot CLI and ZCode. Each run used a clean temp home and a local test model, on 2.0.0-next.15. OMP 18.20.0 ran the same way with ADR-0026, and CI installs the real OMP on Linux, macOS and Windows. For VS Code and Kiro, the check covers the hook setup: VS Code runs the hook command with the documented payload, and kiro-cli 2.35.2 loads all 6 hook entries. CI runs setup, doctor, a hook and an MCP handshake for all 11 agents from the packed package on Linux, macOS and Windows. Record: HOOKS-FIRE.md. Each agent uses its own install command, and ADR-0023 says why.

Your first task with the plugin, step by step: First session. Or paste a prompt from Try it right now.

Try it

Paste one of these prompts into your agent after the install. Each one ran once, live, in Claude Code 2.1.312, and its answer matched the source data (evidence and rerun steps). The headline numbers in Benchmarks use the median of 5 runs. Other agents do the same job with the same tools: ctx_execute, ctx_index and ctx_search. After a prompt, run the stats skill and read the report yourself (See the saving).

On https://react.dev/reference/react/useEffect, show me the example that connects to a chat server and cleans up, and list the rules for the dependency array.

The hook sends WebFetch to the sandbox, and the agent searches the page for the example and the rules. 601,984 B raw, 20,860 B reached the context, 96.5% saved (2026-11-02, 2.0.0-next.26, claude-haiku-5-5).

Fetch https://jsonplaceholder.typicode.com/photos. Which photo has the longest title (give its id and title), and how many photos have a title that contains the word "est"?

The agent runs code over 5,000 JSON records in the sandbox and gets back two answers. 1,039,898 B raw, 393 B reached the context, 99.96% saved (2026-11-02, 2.0.0-next.26, claude-haiku-5-5). The answer is two short facts, so this is the best case of the four.

Here is a Linux syslog sample: https://raw.githubusercontent.com/logpai/loghub/master/Linux/Linux_2k.log. How many authentication failures does it contain, and which 3 remote hosts caused the most?

The agent counts failures in a 1,999-line syslog in the sandbox. 216,485 B raw, 566 B reached the context, 99.7% saved (2026-11-02, 2.0.0-next.26, claude-haiku-5-5).

What are the top 10 stories on the Hacker News front page right now? Give title, points and domain for each.

The hook sends WebFetch to the sandbox, and the agent reads ten stories from the page. 35,087 B raw, 1,501 B reached the context, 95.7% saved (2026-11-02, 2.0.0-next.26, claude-haiku-5-5). The front page changes, so the raw size changes with each run.

See the saving

Run the stats skill right after a prompt. Each agent uses its own command; the skill calls ctx_stats and shows its report unchanged.

Agent

Command

Claude Code

/ctx-stats

Codex

$ctx-stats

Copilot CLI

/ctx-stats

Gemini CLI

Call ctx_stats and show me the report.

Kiro

Call ctx_stats and show me the report.

OMP

/skill:ctx-stats

OpenCode

Call ctx_stats and show me the report.

Pi

/skill:ctx-stats

Qwen Code

Call ctx_stats and show me the report.

VS Code (Copilot)

/ctx-stats

ZCode

Call ctx_stats and show me the report.

The first lines of a real report, after the JSON prompt above ran as one ctx_execute call:

context-mode stats
────────────────────────────────────────────────────────────────────────────────

This session · 1 call · <1m active
  1015.5 KB became 3.7 KB: 1011.8 KB kept out (99.6%)
  Without context-mode |████████████████████████████████████████|  1015.5 KB
  With context-mode    |█░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░|     3.7 KB
  about 259K tokens kept out, 1.3 times a 200K window

Captured 2026-11-03 with 2.0.0-next.35: release-out/server.mjs over MCP in a temp HOME.

Next, the same numbers for 20 real tools, in Use cases.

Use cases

Real tool output from public projects, run through Context Mode. Bench of 2026-11-01 on 2.0.0-next.23 (commit 16a4bcc5), median of 5 runs.

What you do

Tool

Raw

Reaches the context

Saved

Read a browser page

Chrome DevTools MCP take_snapshot, Code Mode

136,514 B

463 B

99.7%

List 40 issues

GitHub CLI gh issue list --json, Code Mode

182,943 B

850 B

99.5%

List 25 pull requests with checks

GitHub CLI gh pr list --json, Code Mode

195,471 B

776 B

99.6%

Look up library docs

Context7 query-docs, Code Mode

6,917 B

473 B

93.2%

Run 1,106 tests

Vitest vitest run --reporter=verbose, Code Mode

62,909 B

429 B

99.3%

Lint a source folder

Biome biome lint --reporter=json, Code Mode

288,523 B

577 B

99.8%

Install a workspace

pnpm pnpm install --frozen-lockfile, Code Mode

14,988 B

199 B

98.7%

Install packages

npm npm install --loglevel http, Code Mode

37,761 B

435 B

98.8%

Read container logs

Docker docker logs, Code Mode

60,374 B

587 B

99.0%

Plan 72 resources

Terraform terraform plan, Code Mode

34,081 B

166 B

99.5%

Read 300 commits with file stats

git git log --stat, Code Mode

164,385 B

268 B

99.8%

Read an 89 KB source file

Read, sent to ctx_execute by the hook

91,362 B

375 B

99.6%

Fetch a 265 KB docs page

WebFetch, sent to ctx_index and ctx_search

271,731 B

439 B

99.8%

Fetch a 214 KB API reply

curl in Bash, replaced after it runs

219,175 B

291 B

99.9%

Print a 187 KB git log in Bash

Bash, output replaced after it runs

191,120 B

204 B

99.9%

Print the same git log through a host tool

Host tool output over the cap, indexed whole

191,120 B

2,842 B

98.5%

Print a 214 KB GitHub API reply through a host tool

Host tool output over the cap, indexed whole

219,175 B

4,060 B

98.1%

Read a 45 KB log file

Read, sent to ctx_execute by the hook

46,216 B

371 B

99.2%

Read a 60 KB source file

Read, sent to ctx_execute by the hook

61,576 B

374 B

99.4%

Debug for a full session: 8 outputs, 3 searches

All hooks and tools

281,261 B

3,266 B

98.8%

Source: bench/results/2026-11-01-darwin-arm64.json, one row per line; SOURCE.md names where each input comes from. Click a task to open its case: the command, the input file with its sha256, and the run command, so you can run the same row on your machine.

The agent gets the part it asks for. The full output stays in the local index, and ctx_search reads it back. Rows with a command run in the sandbox; the agent then finds the rest with search. The Context7 row has 6,917 B of raw output, under the 10 KB line of the headline rule, so the headline leaves it out.

Next, how the bench counts these bytes, in Benchmarks.

Benchmarks

Rule for every headline number in this README: rows with 10 KB or more of raw output, median of 5 runs per row, measured with ctx_bytes against raw_bytes; a category is the median of its rows; 1 KB = 1,024 B. All 36 such rows, the six categories below: the plugin keeps a median of 99.4% of output bytes out of the context. Bench on 2.0.0-next.23 (commit 16a4bcc5 and the local edits that the file records), macOS arm64, 2026-11-01.

Category

What runs

Rows

Median saved

a. Sandbox

ctx_execute on a file or command

19

99.5%

b. Retrieval

ctx_search over the local index

3

77.2%

c. Batch

ctx_batch: many commands, then search

2

93.1%

d. Externalize

Host tool output over the cap, indexed whole

4

98.3%

f. Hook redirect

A big Read, curl or WebFetch sent to the sandbox

7

99.8%

g. Session total

One full scripted session

1

98.8%

Search replies (category b) carry the matching text, so they save 77.2%. Category e (long sessions) is in Longer sessions. The full file lists every row, small ones too: 2026-11-01-darwin-arm64.md. Reproduce with cd bench && node run.mjs. It needs plain Node and runs offline; bench/README.md lists each scenario and its options. Method, data files and rerun steps: Benchmarks guide.

Longer sessions

Say you debug a slow test all afternoon. The agent reads files, runs the suite and fetches docs, and each result goes into the context window. When the window is full, the host compacts it, and from then on the agent sees a short summary plus your next message.

Context Mode keeps the full output of each large tool call in context.db on your machine and gives the agent a small reply. At compaction, a snapshot of at most 2,048 B keeps the session state. After the compaction, the agent calls ctx_resume to read that snapshot back and ctx_search to find any earlier output by keyword.

In the bench, a scripted session of 40 tool outputs and one compaction made a 442,893 B transcript, and the ctx_resume block that brought the agent back was 2,019 B (row e-long-session in the bench file; open the case). In the same bench, one large log of 666,796 B put 3,380 B into the context, and the full log stayed in the index for ctx_search (row d-synth-1m; open the case). How the snapshot picks what to keep: the Long sessions guide and ADR-0009.

Speed

Median of 5 runs, bench of 2026-11-01.

Measure

Result

Hook time over a bare node start, per agent

7.5 ms to 12.3 ms

ctx_search, 20 queries over 8 indexed files

1.62 ms median, 6.70 ms max

ctx_index, 3 real files, 625,649 B

150 ms per MB

Source: rows g-hook-kiro to g-hook-claude-code, g-search-latency and g-index-per-mb of the bench file, with every agent and query. The overhead cases show how each one runs.

Next, what happens inside one tool call, in How it works.

How it works

One tool call, in time order, from the install to a hook error.

  1. Install. Each agent's own install command adds the MCP server, the hooks and the skills ctx-doctor, ctx-stats and ctx-upgrade to that agent's config folder, for example ~/.codex/skills/<name>/SKILL.md. context-mode setup writes or repairs any part the native command leaves out (ADR-0023).

  2. Session start. The SessionStart hook opens the project database. When the session ran before, the agent gets one line that tells it to call ctx_resume (dispatch.ts).

  3. PreToolUse checks. Before a tool runs, the PreToolUse hook checks the call in a fixed order: the user's deny list, WebFetch, curl and wget, inline HTTP and build commands, then a full-file Read over OUTPUT_CAP (16,384 B) and one-file shell dumps such as cat or tail. A Read with a line range runs as usual (decide.ts).

  4. The exact call. When a check matches, the hook stops the call and its reply names the exact ctx_* call to use, for example ctx_execute with {"language":"shell","code":"<the command>","intent":"<what you need>"}. For WebFetch the reply names ctx_index, then ctx_search (read-deny.ts).

  5. The sandbox. This is the Code Mode step: the agent's script runs here, and only what it prints comes back. ctx_execute writes the script to a temp folder and spawns the runtime from an argument list in the project folder. The timeout is EXEC_TIMEOUT_MS (600,000 ms, default and maximum); on timeout it stops the whole process group (taskkill /F /T on Windows). The child gets your environment minus preload variables such as NODE_OPTIONS, and the agent gets what the script prints, at most OUTPUT_CAP (execute.ts). Code Mode. The agent sends a short script as a tool parameter, the script reads the data in the sandbox, and only what it prints goes into the context. Two real calls, in the form the Claude Code terminal shows them:

    ctx_execute on 5,000 JSON records, from the Claude Code 2.1.312 stream of the Try-it run with Context Mode 2.0.0-next.26 (04-api-json.json). The data is 1,039,898 B; the agent gets 254 B.

    ⏺ context-mode - ctx_execute (MCP)(language: "javascript", c…jsonplaceholder.typicode.com/photos\");
      if (!res.ok) throw new Error(\"HTTP \" + res.status);
      const photos = await res.json();
      let longest = null;
      for (const p of photos) {
        if (!longest || p.title.length > longest.title.length) longest = p;
      }
      const withEst = photos.filter(p => p.title.includes(\"est\")).length;
      const ties = photos.filter(p => p.title.length === longest.title.length).map(p => p.id);
      console.log(JSON.stringify({ total: photos.length, longest…nd count of titles containing \"est\"")
      ⎿  exit 0 · 239 B
         {
           "total": 5000,
           "longest": {
             "id": 653,
             "title": "temporibus commodi et fugit doloremque re…rit exercitationem necessitatibus aut",
             "length": 86
           },
           "longestTieIds": [
             653
           ],
           "titlesContainingEst": 892
         }

    ctx_batch on an install log and a long git log, one live MCP call to release-out/server.mjs of Context Mode 2.0.0-next.36 on 2026-11-03 (2026-11-03-ctx-batch.json). The commands print 228,881 B; the agent gets 8,667 B, the parts that match its two queries, and the full output stays in the index.

    ⏺ context-mode - ctx_batch (MCP)(commands: ["cat npm-install… queries: ["cache miss","feat commit"])
      ⎿  ## cache miss
         js.org/@esbuild%2fwin32-x64 282ms (cache miss)
         npm http fetch GET 200 https://registry.npmjs.org/@esbuild%2flinux-s390x 79ms (cache miss)
         npm http fetch GET 200 https://registry.npmjs.org/@esbuild%2fopenbsd-x64 101ms (cache miss)
         npm http fetch GET 200 https://registry.npmjs.org/@esbuild%2fwin32-arm64 99ms (cache miss)
         npm http fetch GET 200 https://registry.npmjs.org/@esbuild%2fdarwin-arm64 99ms (cache miss)
         … +95 lines (ctrl+o to expand)
  6. Indexing. The full output goes into the index in chunks of at most MAX_CHUNK_BYTES (4,096 B). A Markdown chunk stays inside one heading section, and a code block that fits stays whole. Each chunk goes into two SQLite FTS5 tables: a Porter-stemmed table for words and a trigram table for paths, names and parts of words (fts.ts).

    ctx_index on a copy of this repo's docs/how-it-works.md, one live MCP call to release-out/server.mjs of Context Mode 2.0.0-next.36 on 2026-11-03 (calls.json). The file is 11,577 B; the agent gets 64 B.

    ⏺ context-mode - ctx_index (MCP)(path: "how-it-works.md")
      ⎿  Indexed 1 source(s), 17 chunk(s) from s1. Label: how-it-works.md
  7. Search. ctx_search ranks each table with BM25 and joins the two lists with Reciprocal Rank Fusion: a hit scores 1/(RRF_K + rank), with RRF_K = 60. Before it returns an empty list, it tries a spelling fix from the indexed words. Search covers this session, this worktree (the default) or the whole project, and each snippet is at most SNIPPET_MAX_BYTES (800 B) (search.ts).

    ctx_search on that index, in the same run, in a project that also holds an AGENTS.md. One search reads the index and every agent's memory files. The agent gets 1,268 B: one content hit and one memory hit, each with its source and heading. For memory shared across machines, agents and sessions, see Context Mode Gateway Memory.

    ⏺ context-mode - ctx_search (MCP)(queries: ["compaction state block"])
      ⎿  Query "compaction state block":
         how-it-works.md #14: ### Session continuity
         
         The hooks log each prompt, edit, command, tool error, i…e same rows always give the same bytes.
         memory AGENTS.md #0: # Agent rules
         … +10 lines (ctrl+o to expand)
  8. Run, then replace. On hosts that run the tool first, the PostToolUse hook checks the result. Output over REPLACE_TRIGGER_BYTES (4,096 B) goes into the index whole, and the agent gets the best-matching chunks and one ctx_search line for the rest (dispatch.ts).

  9. Journal and compaction. The hooks log each prompt, edit, command, tool error and decision in a journal. After a compaction the agent gets a state block of at most STATE_BUDGET (2,048 B), and each long line becomes a ctx_resume(seq) pointer to the full row; one ctx_resume reply is at most RESUME_CAP (2,048 B). On the next prompt, a prompt-match block brings back the journal rows that match it by BM25 (ADR-0009).

    ctx_resume with empty parameters returns the state block, in the same run: 209 B.

    ⏺ context-mode - ctx_resume (MCP)()
      ⎿  <context_mode_state session="mcp:85e67fa2-28ef-45eb-9d5b-a967129b74c0" upto="2">
         indexed how-it-works.md
         source s1: how-it-works.md (11536 B)
         more: ctx_search(queries:[], scope:"session")
         </context_mode_state>
  10. The database. Each project has one context.db under ~/.context-mode, in Node's built-in node:sqlite. The database runs in WAL mode and runs a PASSIVE checkpoint on close. A new database gets built under a temp name and moves into place in one atomic step, and the busy timeout is DB_TIMEOUT_MS (5,000 ms) (db.ts).

  11. A hook error. A hook always ends with exit 0. On any error it gives the host's allow reply and writes one log line to ~/.context-mode/hook-errors.log, so your agent keeps working. A spawned hook has HOOK_TIMEOUT_MS (5,000 ms) (fallback.ts).

Every number with its reason: How it works. The rules the hook follows: ADR-0008. Next, the Tools the agent calls.

Tools

Tool

What it does

ctx_execute

Runs code or a shell command in a sandbox; returns what it prints

ctx_batch

Runs many commands, indexes the output, and searches it in one call

ctx_index

Indexes a file, a text or a URL into the project database

ctx_search

Searches the local index; returns the matching parts

ctx_resume

Returns the plugin's local state block (at most 2,048 B), or one full row by seq

ctx_stats

Shows how much output stayed out of context

ctx_doctor

Checks the install; each check prints PASS or a fix command

ctx_upgrade

Installs, upgrades or uninstalls; replaces old contextmode entries

ctx_purge

Deletes indexed data by session, source or project

Three more real calls, from the same live run of Context Mode 2.0.0-next.36 on 2026-11-03 (calls.json).

ctx_stats: the agent gets 3,097 B, the first line is the summary.

⏺ context-mode - ctx_stats (MCP)()
  ⎿  ctx_stats: 82.5% kept out all-time, 82.5% this session.
     … +59 lines (ctrl+o to expand)

ctx_doctor after context-mode setup --agents claude-code: 862 B, one row per check.

⏺ context-mode - ctx_doctor (MCP)()
  ⎿  Environment: context-mode 2.0.0-next.36 (8facf27), agent claude-code 0 (client r20-live)
     Environment: darwin-arm64, Node v26.8.1, shell /bin/zsh
     | Check | Status | Detail | Fix |
     |---|---|---|---|
     | D1 version | PASS | 2.0.0-next.36 (8facf27): ahead of next on npm (2.0.0-next.34) |  |
     | D2 node | PASS | ok |  |
     … +11 lines (ctrl+o to expand)

ctx_purge deletes one source: 66 B.

⏺ context-mode - ctx_purge (MCP)(source: "how-it-works.md")
  ⎿  ctx_purge: deleted 18 rows (scope: source). This cannot be undone.

Full reference: Tools.

Commands in the agent: /ctx-stats shows the ctx_stats result, /ctx-doctor runs ctx_doctor (send its report with a defect report), and /ctx-upgrade runs ctx_upgrade, then ctx_doctor.

Security and privacy

  • The plugin keeps all its data on your machine, one database per project in ~/.context-mode, and ctx_purge deletes it by session, source or project. Read the Security guide.

  • The plugin has zero telemetry: usage data, events and crash reports stay on your machine. It uses the network for four calls, each one started by you or your agent:

Call

When

File

Fetch of a URL, for ctx_index

Your agent asks for a page. The fetch blocks private and link-local addresses, checks each address the host name points to, connects to the address it checked, and checks again at each redirect.

fetch.ts

One read of the npm dist-tags, with git ls-remote of the release tags as the fallback

You run context-mode doctor, to check for a new version

doctor-env.ts

npm i -g context-mode, or gh issue create

You run context-mode upgrade, or you press a key to send the debug report

commands.ts

Your agent's own install command, npm or git

You install or update

Your agent

  • The guard test D-PRIVACY-NET-1 in tests/0008-enforcement.test.mjs scans the source for HTTP clients, sockets, DNS lookups, git network calls and gh or npm spawns. It flags each new call site, so this table stays true.

  • Doctor and the debug report remove secrets by their structure before they print.

Code from ctx_execute runs on your host as your user, with your files, your network and your environment except the preload variables, so keep your agent's own permission prompts on. Here "sandbox" means the output stays in the sandbox and out of the context. What the plugin stores, where, and how to delete it: Privacy.

Context Mode Gateway

The plugin works on tool output, on your machine, and it is complete on its own. Context Mode Gateway is a separate hosted service with its own account and subscription. It works on every request, across all your agents. Its 11 features, the full list: Context Saving, Recall, Memory, Skills, Search, Session resume, Active Context, Instructions, Protect, Prompt caching, Live and Analytics. Add the Gateway when you want savings on every request, memory across sessions and agents, or cost reports for a team. For one project on one machine, the plugin covers the job. The two work together: Gateway guide. Gateway Recall, Search and Session resume run on the Gateway server, across sessions and agents. The plugin's ctx_search and ctx_resume read the local context.db of one project. With the Gateway on, your requests pass through its server; the plugin keeps its own data on your machine.

Product

Runs

Scope

Billing

Context Mode plugin

On your machine

One project, local context.db

Free

Context Mode Gateway

Hosted service

All sessions and agents

Subscription, separate from the plugin

Set up the Gateway with one command:

npx @context-mode/cli

npx @context-mode/cli signs you in to your Gateway account, which has its own subscription, and writes one config change per agent. Apps: Claude Code, Claude Desktop, Codex, Codex Desktop, T3 Code, OpenCode and Pi. The Gateway CLI needs Node 20.12 or later, from its npm engines field.

Live. One session as it runs: money saved this session, bytes kept out per request, history kept and sent, and a feed of each request with its cost and saving. Click the image for the full video.

Docs: How saving works

Numbers in these screens come from the author's own account over 30 days (Oct 3 - Nov 1, 2026), so read them as one account's usage. Click an image for the full size.

Three prompts to try with the Gateway on. The line under each one names the feature it shows.

Move this project from Jest to Vitest. Convert every test file, fix the config, run the suite after each batch, and tell me what is left.

Memory keeps the plan and the file list across a long job.

Pick up where we stopped yesterday on the Jest to Vitest move. What is done, what is left, and which test was failing?

Recall brings back yesterday's work in a fresh session.

Which date library did we choose on Tuesday, and why did we reject the other two?

Search finds the answer in past sessions.

Source: contextmode.ai/docs:

Feature

What it does

What you get

Details

Context Saving

Folds old tool output, runs data work as code (Code Mode), trims tool schemas, places cache marks

A tool output over 2,048 B becomes a short pointer, with the full copy archived

Saves the most on sessions with many tool calls

Recall

The model sees recent work and an index of older work; the agent searches the archive for the rest

Gateway, Claude Code, Opus 5.5, 200K window, 2026-10-04: mean cost per session $4.8627 to $3.0353 (-37.6% in dollars, ab.json below)

Cost measured in Claude Code on Claude Opus 5.5

Memory

Each fact keeps two times: when it was true and when the Gateway learned it

Claude Code and Codex on one account share one store, tested end to end

The memory block goes into the first request of a new session

Skills

A skill catalog on the Gateway for your account; it loads the skill that matches your message

The matching skill loads on the message that needs it

Import skills from GitHub; a loaded skill joins the prompt in full

Search

Searches the prompts, commands and tool output that passed through, in the console and as an agent tool

Find any past prompt, command or tool output, from the console or your agent

The console searches the first 4,000 characters of each tool result

Session resume

One command rebuilds a conversation as a native Claude Code or Codex session

Continue a conversation in Claude Code or Codex

Restores the text, tool names and a summary; the command is valid for 15 minutes

Active Context

Sets how much a thread carries before its oldest parts move to the archive: 200K, 500K or 1M tokens

Pick the window size per thread

An image older than the kept part becomes a short text note

Instructions

One text of at most 4,096 characters that tells every agent how to answer you

Read from the provider cache after the first request

Sets how the agent writes to you

Protect

A tool-call firewall: checks shell commands, file reads and writes, web fetches and tool-server calls by what they do

Each call checked before it runs

Reads the call itself, so keep your agent's sandbox on for script files and paths built at run time; opt-in

Prompt caching

Keeps requests cache-friendly: cache marks on the fixed part of the prompt and a cache for fetched pages

Inside the measured costs above

Uses up to 4 cache marks, the provider's maximum

Live and Analytics

Shows each request and what the Gateway saved, as it happens and over time

See each request and its saving, live and over time

Covers every request that passes through the Gateway

Gateway only: its cost against Claude Code alone, Claude Opus 5.5, mean dollars per session at list price. The plugin numbers are in Benchmarks.

  • Six short tasks (2026-09-29): 24.5% fewer dollars pooled over 176 sessions (21% to 29% per task). v3-opus55-short.json

  • Long Claude Code session, 200K window (2026-10-03/04): mean cost per session $4.8627 to $3.0353 (-37.6% in dollars, mean of 3 pairs). ab.json

  • We fixed the rules before each run: contextmode.ai/docs/benchmarks.

The Gateway routes to Anthropic and OpenAI. The Gateway sends your requests to its server on the way to the model. It keeps a searchable record of your sessions for Recall and Search, and masks secrets and e-mail addresses when it reads archived output back. Its privacy page is at contextmode.ai/privacy. Keep your agent's own sandbox on next to Protect.

License

Elastic License 2.0 (ELv2, the LICENSE file), a source-available license. You may use, copy, change and share the code, and use it at work, free of charge. The license keeps three uses for the licensor:

  1. Offering the software to others as a hosted or managed service that gives them a large part of its features.

  2. Moving, changing, turning off or getting around the license key functions, or removing or hiding what the license key protects.

  3. Changing, removing or hiding the licensor's license, copyright or other notices.

We chose ELv2 so the source stays open to all and the hosted service stays with the project.

Community

Report a defect on Discord with the full /ctx-doctor report, and send ideas there too. The maintainer writes every change; CONTRIBUTING.md shows how an agent files an issue.

Videos. Better Stack on Context Mode. Make your own video and share it on Discord; we add videos over 10K views here.

Available Tools

11 tools
ctx_batch_executeBatch Execute & SearchA
Destructive

Run multiple commands in ONE call. Every command's output is auto-indexed into the knowledge base; if you also pass queries, the matching sections come back in the same round trip so a follow-up search call is not needed.

Concurrency parallelizes the FETCH phase (run-the-commands). The DERIVATION phase — turning raw output into an answer — still belongs in code: add a processing command that consumes the indexed output and prints only the answer, so the raw bytes never enter your conversation (Think-in-Code, same principle as the sandbox tool).

WHEN:

  • You have 3+ related commands you would otherwise run sequentially (multi-issue lookups, git log + git diff + git blame, multi-file reads, multi-region cloud queries)

  • You want to gather AND query in one round trip — pass queries so the matching sections come back inline

  • You want to parallelize I/O-bound work — pass concurrency 2-8 (network calls, gh CLI, cloud APIs, multi-repo git reads)

  • The combined output is large enough that piping it through ctx_search later would itself be expensive — let auto-index + inline queries do both in one shot

WHEN NOT:

  • Single command with no follow-up query — run it in the sandbox tool directly

  • CPU-bound or stateful commands — keep concurrency at 1 (npm test, build, lint, port-binding servers, lock-file holders, anything that races on the same resource)

RETURNS: Auto-indexed section list per command label, plus top matches per query (when queries is passed). Raw output is NOT echoed in full — only the matched windows. Concurrency>1 switches each command to its own per-command timeout (no shared budget); concurrency=1 preserves the legacy shared-budget cascading-skip-on-timeout path. Use 4-8 for I/O-bound batches; keep at 1 for CPU work or shared-state commands; lower the value when target hosts enforce per-IP rate limits.

EXAMPLE: ctx_batch_execute( commands: [ {label: "issue 1", command: "gh issue view 1"}, {label: "issue 2", command: "gh issue view 2"}, {label: "summarize", command: "echo done"} ], queries: ["root cause", "proposed fix"], concurrency: 2 )

Inputs: queries (required), commands (required), cwd, timeout, +2 more.

ctx_doctorRun DiagnosticsA
Read-onlyIdempotent

Diagnose context-mode installation. Runs all checks server-side and returns a plain-text status report with [OK]/[FAIL]/[WARN] prefixes (renderer-safe across MCP clients). No CLI execution needed.

Inputs: no named parameters.

ctx_executeRun code in a sandbox (executes the supplied code)A
Destructive

Run code in a sandboxed subprocess. Languages: javascript, shell, typescript, python, perl.

Think-in-Code — the core philosophy: the bytes your code processes never enter your conversation memory; only what you console.log() does. Reading a 700 KB log directly means 700 KB of your remaining reasoning capacity gets spent on raw bytes. Running code over that same log in this sandbox and printing a 3 KB summary leaves you with 697 KB of capacity for the actual work.

Concrete shape — analyze 47 source files without reading any of them: ctx_execute(language: "javascript", code: const fs = require('fs'); const files = fs.readdirSync('src').filter(f => f.endsWith('.ts')); files.forEach(f => { const lines = fs.readFileSync('src/'+f,'utf8').split('\\n').length; console.log(f + ': ' + lines + ' lines'); }); ) // 47 files analyzed, 15,314 LoC summarized — output ~3.6 KB instead of 47 Read() calls = ~700 KB.

WHEN:

  • You intend to derive an answer FROM data (filter, count, aggregate, parse, compare, transform) — do the derivation in code and print only the answer

  • Output shape or size cannot be predicted before execution (recursive finds, repo-wide greps, list endpoints, query results, log scans)

  • You would otherwise read raw output and then mentally compute — that compute belongs here, in code, where its inputs stay out of your conversation

  • You need to keep a long-running process alive (dev server, watcher, daemon) — pass background: true to detach on timeout instead of killing the process

  • The output may legitimately be large but you only want recall-by-topic later — pass an intent string; outputs over ~5KB are auto-indexed into the knowledge base and only the section titles + previews come back, retrievable via ctx_search

WHEN NOT:

  • Single observational command whose entire short output you intend to consume verbatim (whoami, pwd, git status on a clean tree) — Bash is simpler

  • File mutations (Edit/Write) or navigation (cd/ls) — Bash is the right surface

  • You already know the output is one short fixed line and you want to read it as-is

RETURNS: Only what your code prints. Wrap risky calls in try/catch — uncaught errors go to stderr and may leak more than intended. When intent is set and output exceeds the auto-index threshold, the response carries searchable section titles + previews instead of the raw stdout; use ctx_search(queries: [...]) to drill into specific sections.

EXAMPLE: ctx_execute(language: "javascript", code: "const out = require('child_process').execSync('npm test', {encoding:'utf8', stdio:['ignore','pipe','pipe']}); console.log(out.split('\n').filter(l => /(FAIL|✗|×|Error:|Tests +.*(failed|passed))/i.test(l)).slice(0, 60).join('\n'))") EXAMPLE: ctx_execute(language: "javascript", code: "const out = require('child_process').execSync('gh issue list --json number,title --limit 100', {encoding:'utf8'}); const hooks = JSON.parse(out).filter(i => /hook|routing/i.test(i.title)); console.log(${hooks.length} hook-related issues)")

Inputs: code (required), language (required), cwd, intent, +2 more.

ctx_execute_fileRun code over a file (executes code, reads the given path)A
Destructive

Read a file into a sandboxed FILE_CONTENT variable and run code over it. Only what you console.log() enters your conversation — the file bytes stay in the sandbox.

Think-in-Code applied to file-level analysis: Reading the whole file means every byte enters your conversation memory and costs reasoning capacity for the rest of the session. Running code over it here lets you keep the raw bytes out and only the derived answer in. Same principle as ctx_execute, scoped to one named file via the FILE_CONTENT variable.

WHEN:

  • You want to KNOW SOMETHING ABOUT a file (line count, matches of a pattern, parsed structure, statistical aggregate) without needing to SEE all of it

  • The file is structured (CSV, JSON, log, code) and a code-level derivation is cheaper than reading verbatim

  • The file is large enough that reading the full content would burn meaningful conversation memory you need for the actual work

  • The derivation may itself produce a large output you want recall-by-topic on later — pass an intent string; outputs over ~5KB are auto-indexed and only matching sections come back, retrievable via ctx_search

WHEN NOT:

  • You intend to EDIT the file — use Read so the subsequent Edit can match the exact text

  • You only need one specific line and you know its offset — Read with offset/limit is the simplest path

  • The file is small AND you will consume all of it for understanding/editing — Read directly

RETURNS: Only what your code prints. The FILE_CONTENT variable holds the raw bytes inside the sandbox; nothing else leaves. When intent is set and output exceeds the auto-index threshold, the response carries searchable section titles + previews instead of the raw stdout.

EXAMPLE: ctx_execute_file(path: "huge.log", language: "javascript", code: "const errs = FILE_CONTENT.split('\n').filter(l => /ERROR|FATAL/.test(l)); console.log(${errs.length} error lines); console.log(errs.slice(-5).join('\n'))") EXAMPLE: ctx_execute_file(path: "data.csv", language: "javascript", code: "const rows = FILE_CONTENT.split('\n'); console.log(rows: ${rows.length - 1}, header: ${rows[0]})")

Inputs: code (required), path (required), language (required), intent, +1 more.

ctx_fetch_and_indexFetch & Index URL(s)A

Fetches URL content, converts HTML to markdown (JSON is chunked by key paths, plain text indexed directly), persists it in a searchable knowledge base, and returns a small preview window per source. The raw page bytes never enter your conversation — they live in storage and you retrieve any section on-demand via ctx_search.

Caching: every fetch is cached on disk and reused for repeat calls within the TTL window. The default TTL is 24 hours; override per-call with the ttl parameter (milliseconds, ttl: 0 bypasses cache like force: true). Stored content older than 14 days is cleaned up on startup.

WHEN:

  • You need web content (docs, changelogs, API references, spec pages) and the raw page bytes should NOT enter your conversation

  • Multi-URL research (library evaluation, migration scans, doc comparisons): pass the requests array and a concurrency value 2-8 for parallel I/O

  • You want repeat lookups against the same URL to be cheap (TTL cache hits return only a hint, no re-fetch)

  • You want a long-lived cache window (override ttl upward for stable specs) or a guaranteed-fresh fetch (ttl: 0 or force: true)

WHEN NOT:

  • You already have the content locally — store it via the inline index tool

  • The page is SPA-rendered (JavaScript-required to materialize content) — this is a plain HTTP fetch, no headless browser

RETURNS: Per-source preview windows extracted around indexable headings plus indexing metadata (chunk counts, source labels, cache state). Raw content is NOT echoed back — retrieve any section on-demand via ctx_search(source: ""). Concurrency parallelizes the fetch phase up to your chosen value (capped by the host's logical CPU count); the FTS5 write phase always runs serially because SQLite is a single-writer store. Net latency = max(fetch latency across the pool) + sum(per-source index write time). Cache hits skip both phases and return a small freshness hint instead of re-fetching. Use 4-8 for stable I/O-bound batches; lower the value when the target host enforces a per-IP rate limit you cannot raise.

EXAMPLE: ctx_fetch_and_index( requests: [{url: "https://react.dev/...", source: "react"}, {url: "https://vuejs.org/...", source: "vue"}], concurrency: 5 )

Inputs: ttl, url, force, source, +2 more.

ctx_indexIndex ContentA

Store content in a searchable knowledge base (BM25 over FTS5). Splits markdown by headings, keeps code blocks intact, and persists the raw chunks. The full content stays in storage — retrieve any section on-demand via ctx_search; nothing is summarized or truncated.

WHEN:

  • Documentation from Context7, Skills, or MCP tools (API docs, framework guides, code examples)

  • API references (endpoint details, parameter specs, response schemas)

  • MCP tools/list output (exact tool signatures and descriptions)

  • Skill prompts and instructions that are too large to keep verbatim in conversation

  • README files, migration guides, changelog entries

  • Any content with code examples you may need to reference precisely later

WHEN NOT:

  • Log files, test output, CSV, or build output — use ctx_execute_file, which processes in-sandbox without persisting bytes

  • Single-use ephemeral content you will not query later — keep it inline if it fits, or ctx_execute_file it

RETURNS: Indexing metadata: chunk counts (total, code-bearing), source label, and the exact ctx_search call shape to query the indexed content. Raw content is NOT echoed back — it lives in storage, retrievable via ctx_search(source: ""). When path is provided, a content hash is stored so ctx_search results auto-flag staleness on future calls.

EXAMPLE: ctx_index(content: "# React useEffect\n\nThe Effect Hook lets you ...", source: "react-useeffect-docs") EXAMPLE: ctx_index(path: "/path/to/large-spec.md", source: "openapi-v2-spec")

Inputs: path, source, content, exclude, +6 more.

ctx_insightOpen Insight DashboardA
Idempotent

Opens the context-mode Insight dashboard (https://context-mode.com/insight) in your default browser — a dashboard launcher for the hosted analytics layer, not a Q&A engine. Insight surfaces per-engineer productive rate, retry waste, blocker detection, and role-narrowed views for CTO, EM, IC, CISO, FinOps, and DevOps. For natural-language queries over your indexed content, use ctx_search.

Inputs: no named parameters.

ctx_purgePurge Knowledge BaseA
DestructiveIdempotent

DESTRUCTIVE: permanently delete indexed content. Cannot be undone. Requires confirm:true and exactly one scope.

WHEN:

  • User explicitly asks to clear a specific session ('purge this session', 'wipe this conversation')

  • User explicitly asks to reset the whole project ('reset everything', 'wipe the knowledge base')

WHEN NOT:

  • User says 'reset', 'clear', or 'wipe' without naming a scope -> ask which scope before calling

  • User wants to free memory or improve performance -> recommend ctx_stats first, do not purge

SCOPES (pass exactly one):

  • Per-session: ctx_purge(confirm: true, sessionId: "") deletes that session's events (auto-captured decisions, errors, plans, user prompts, rejected approaches, etc.) and per-session FTS5 chunks; sibling sessions and stats file are preserved.

  • Per-project: ctx_purge(confirm: true, scope: "project") wipes FTS5 knowledge base, every session DB row, events markdown, and resets the stats file. Use ctx_stats first to preview category counts before purging.

CONTRACT:

  • confirm:true is required; confirm:false returns 'purge cancelled'.

  • sessionId and scope:'project' together return 'ambiguous - pick one'.

  • scope:'session' without sessionId throws (sessionId required).

  • Bare {confirm:true} is deprecated: maps to scope:'project' with a stderr warning; will hard-error in a future major.

RETURNS: A summary of removed rows + the resolved scope.

EXAMPLE: ctx_purge(confirm: true, sessionId: "7c8a-1234-5678-9abc-def012345678") EXAMPLE: ctx_purge(confirm: true, scope: "project")

Inputs: confirm (required), scope, sessionId.

ctx_statsSession StatisticsA
Read-onlyIdempotent

Returns context consumption statistics for the current session. Shows total bytes returned to context, breakdown by tool, call counts, estimated token usage, and context savings ratio.

Inputs: no named parameters.

ctx_upgradeUpgrade PluginA
Idempotent

Upgrade context-mode to the latest version. Returns a shell command to execute. You MUST run the returned command using your shell tool (Bash, shell_execute, run_in_terminal, etc.) and display the output as a checklist. Tell the user to restart their session after upgrade.

Inputs: no named parameters.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.4/5.0

    Scored across 11 tools

    Disambiguation4/5

    ctx_execute, ctx_execute_file, and ctx_batch_execute share an execution surface, but each has a clearly scoped role (single code block vs file-scoped processing vs batched commands with auto-indexing) reinforced by detailed WHEN/WHEN NOT guidance. ctx_index and ctx_fetch_and_index are similarly distinct (inline content vs URL fetching). No two tools are truly interchangeable.

    Naming Consistency5/5

    Every tool uses the ctx_ prefix with lowercase snake_case, and most follow a verb pattern (ctx_execute, ctx_index, ctx_search, ctx_purge, ctx_upgrade) or a noun for informational tools (ctx_stats, ctx_doctor, ctx_insight). Fetch-and-index and batch-execute combine verbs in a predictable, readable way.

    Tool Count5/5

    11 tools cover the server's scope — sandboxed execution, knowledge-base indexing/retrieval, maintenance (doctor, upgrade), statistics, and destructive cleanup — without redundancy. Each tool has a distinct purpose, and the count is within the ideal range for a feature-rich but focused MCP server.

    Completeness4/5

    The main lifecycle is covered: capture (ctx_execute/ctx_execute_file/ctx_batch_execute/ctx_fetch_and_index), store (ctx_index), retrieve (ctx_search), and destroy (ctx_purge). Minor gaps remain: there is no per-source or per-chunk deletion/update tool, and no explicit way to list all indexed sources beyond stats counts.

    Maintenance

    ActivityActive
    ResponsivenessUsually handled

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      Local-first MCP server that gives any AI coding agent per-project memory, workflow intelligence, and always-on, lossless token & context optimization.
      37
      31 npm
      6
      MIT
    • A
      license
      C
      quality
      C
      maintenance
      An MCP server that filters and compresses context by 80-90% before sending to an LLM, using code knowledge graphs and compression.
      20
      69 PyPI
      3
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      An MCP server that provides a persistent sandbox for AI coding agents to explore codebases server-side, returning only compact summaries to reduce context consumption.
      3
      9
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Provides reversible context compression for AI agents, reducing token usage while preserving the ability to retrieve original content, and serves as an MCP server for integration with tools like GitHub Copilot and Claude Code.
      3
      1
      Apache 2.0