@contextstream/mcp-server 1.0.34 → 1.0.36

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +33 -34
  2. package/package.json +1 -1
  3. package/server.json +3 -3
package/README.md CHANGED
@@ -6,13 +6,22 @@
6
6
 
7
7
  <h1 align="center">ContextStream MCP Server</h1>
8
8
 
9
- <p align="center"><strong>Persistent memory. Millisecond code search.</strong></p>
9
+ <p align="center"><strong>The context platform for AI agents.</strong></p>
10
10
 
11
11
  ## Your agent knows code. Give it the decisions behind yours.
12
12
 
13
13
  Connect your code, docs, and conversations so your agents can find the right
14
14
  files, recall saved decisions, and build on past work across sessions and tools.
15
15
 
16
+ <a href="https://contextstream.io/benchmarks">
17
+ <picture>
18
+ <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/contextstream/mcp-server/main/assets/benchmarks-dark.png">
19
+ <img src="https://raw.githubusercontent.com/contextstream/mcp-server/main/assets/benchmarks-light.png" width="100%" alt="ContextStream benchmarks: LongMemEval-S 96.8% (484 of 500), LoCoMo 95.7%, code search 98.77% best answer in the top 10, agent tasks 96% (23 of 24, up from 58% without it)">
20
+ </picture>
21
+ </a>
22
+
23
+ The numbers, their sources, and what they do not claim are under [Benchmarks](#benchmarks).
24
+
16
25
  <a id="install-in-30-seconds"></a>
17
26
 
18
27
  ## Start free with MCP
@@ -55,39 +64,29 @@ Then ask it to save a real decision and its reason. Start a new session and
55
64
  retrieve that decision without explaining it again. Indexing supplies code
56
65
  context; it cannot recover every undocumented decision.
57
66
 
58
- ## Slipstream agents
59
-
60
- When your workspace has Slipstream, the server adds one tool, `slipstream`, that hands a
61
- whole job to a ContextStream specialist agent working with your workspace's context:
62
- review a pull request against your team's decisions, write release notes, check docs
63
- against the code, tend memory, write a repository guide, read an image, make a PDF, or
64
- digest your mail (it never sends). Each agent is priced in credits per run and charged
65
- only when the run succeeds; declined, failed, cancelled and timed-out runs are free.
66
-
67
- - **Enabling.** A workspace owner or admin turns agents on in the workspace's Slipstream
68
- settings. The tool is listed only for a workspace with at least one agent turned on, so
69
- everywhere else it costs no prompt. Its description names each agent turned on, its
70
- price and the input it needs, so you can just ask: "Which Slipstream agents can I use
71
- here?" or "Have Slipstream review PR 42 in acme/api."
72
- - **Workspace.** The local server lists it for the workspace of your session (or of
73
- `CONTEXTSTREAM_WORKSPACE_ID`, or of the folder's mapping) and tells clients that follow
74
- `notifications/tools/list_changed`, such as Claude Code, when that changes. A hosted
75
- connection lists it for the workspace its `X-ContextStream-Workspace-Id` header names.
76
- - **Prices.** A priced agent first answers with its price; your agent shows it to you and
77
- runs the job once you accept. To let runs up to a price start without that step, set an
78
- auto-accept cap: `CONTEXTSTREAM_SLIPSTREAM_AUTO_ACCEPT_MAX_CREDITS=500` in the local
79
- server's environment (your MCP config's `env`), or the header
80
- `X-ContextStream-Slipstream-Auto-Accept-Max-Credits: 500` on a hosted connection. Unset
81
- or `0` always asks; the hosted gateway never sets a cap for you.
82
- - **Runs.** A run waits up to 25 seconds for its result; a longer one returns its task id
83
- for your agent to follow. A retried call never starts a second run.
84
- - **Terminal.** The same agents run from a shell with `ss run <agent>` (alias `csa`).
85
-
86
- ## Proof and data controls
87
-
88
- **95 ms median code search in our published benchmark.** This is the measured
89
- successful-response median for the disclosed configuration, not a latency guarantee.
90
- [See results, methodology, and limitations](https://contextstream.io/benchmarks).
67
+ ## Benchmarks
68
+
69
+ Each benchmark runs through the production API, is registered before its results
70
+ are read, and is published with its method, confidence interval, and losses.
71
+
72
+ | Benchmark | ContextStream | Compared with |
73
+ |---|---|---|
74
+ | **LongMemEval-S**, all 500 questions | **96.8%** (484/500); 96.0% on the half held out from tuning | Above Mastra (94.87%) and supermemory (85.4%) on the same official GPT-4o judge. Level with agentmemory's self-reported 96.2%, which is inside our 95% interval |
75
+ | **LoCoMo**, all 1,540 questions | **95.7%** (1474/1540); 95.8% held out; 97.8% on the audited answer key | Above every system in the EverMemOS paper's same-judge table: EverMemOS 93.05%, Zep 85.22%, MemOS 80.76%, MemU 66.67%, Mem0 64.20% |
76
+ | **Code search**, 177 judged developer queries | **98.77%** best answer in the top 10; **100%** a useful file in the top 10 | Embedding-only search 72.84% and ripgrep 42.59% on the same queries and judgments |
77
+ | **Agent tasks**, 8 tasks × 3 trials | **23/24** passed (96%); 9/9 on tasks that need team knowledge | 14/24 (58%) and 0/9 with the repository alone |
78
+
79
+ **61 ms median server-side search time** (127 ms p95) in the October 27 production
80
+ run. End to end it was 223 ms, because the client ran in a different region from the
81
+ one serving the workspace. Your latency depends on your distance from that region.
82
+
83
+ Competitor figures are as each vendor or the EverMemOS paper published them; we did
84
+ not re-run them, and readers and judges differ across rows. The code-search baselines
85
+ come from the August 1 run on the same queries. The agent-task result uses one
86
+ repository and one agent model, so treat it as directional.
87
+ [Results, methodology, and limitations](https://contextstream.io/benchmarks).
88
+
89
+ ## Data controls
91
90
 
92
91
  **Scoped access. Traceable sources. Configurable capture.**
93
92
  Indexing sends eligible source contents to ContextStream for hosted search.
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "@contextstream/mcp-server",
3
3
  "mcpName": "io.github.contextstream/mcp-server",
4
- "version": "1.0.34",
4
+ "version": "1.0.36",
5
5
  "description": "Verified npm launcher for the open-source ContextStream Rust MCP server",
6
6
  "type": "module",
7
7
  "license": "MIT",
package/server.json CHANGED
@@ -2,8 +2,8 @@
2
2
  "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
3
3
  "name": "io.github.contextstream/mcp-server",
4
4
  "title": "ContextStream MCP Server",
5
- "description": "Project memory, semantic code search, and grounded agent context.",
6
- "version": "1.0.34",
5
+ "description": "The context platform for AI agents: code search, team decisions, and grounded project context.",
6
+ "version": "1.0.36",
7
7
  "repository": {
8
8
  "url": "https://github.com/contextstream/mcp-server",
9
9
  "source": "github"
@@ -20,7 +20,7 @@
20
20
  "registryType": "npm",
21
21
  "registryBaseUrl": "https://registry.npmjs.org",
22
22
  "identifier": "@contextstream/mcp-server",
23
- "version": "1.0.34",
23
+ "version": "1.0.36",
24
24
  "transport": {
25
25
  "type": "stdio"
26
26
  },